On 03/11/2026 20:25, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe
thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work. The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
Minis (and later superminis, like the VAX) had it all over mainframes
in that regard.
Why? I don't need to imagine, I did it all the time. Its not a problem.
The editor generates a screen full of data.
The controller handles most of the character editing so you change the
local buffer which does see every key stroke.
At some point you send the modified screen back to the host.
Personally I prefer the XEDIT the VM/CMS editor to the VMS one, but then
I am more familiar with it.
[...]
Recalling at the time that interactive used was typically form-based
(e.g. airline reservation systems, bank teller terminals, etc) and
the block mode terminals of the day typically transferred the contents
of the screen (or the form fields on the screen) in a single transmission.
That's not quite true. First of all old IBM COBOL and RPG is well
supported on modern -|P based hardware. and second mainframes are still
in use where enormous numbers of I/O operations are called for.
"Mainframes remain the invisible backbone of the global economy in the
21st century, handling 90% of credit card transactions, powering 92 of
the top 100 banks, and managing 68% of production IT workloads globally while accounting for just a fraction of total IT spending."
Mainframes remain cost effective alternatives for very large scale low complexcity transactions
[...]
-a The age of the "mainframe" may be coming to an
-a end alas. 'Cluster'/AI solutions are going to
-a replace them for most uses. However science/govt
-a still have uses for big ultra-powerful boxes
-a for awhile anyway.
SOME advantage in keeping
-a access 'local' too - as opposed to spread all
-a over the highly-vulnerable net.
[...]
I know at least one guy working on IBM mainframes, and he tells me they
are in high need of good assembler programmers.
[...]
Mainframes remain cost effective alternatives for very large scale low
complexcity transactions
The above is mostly IBM advertising. When I looked at mainframe
technical data IBM put there some impressive numbers. But I think
that is is rather rare that IBM mainframes are best fit for the
problem. AFAICS main IBM technical advantage was in their
mutichip module packing technology: they put there several chips
with fast interconnect and efficiently removed heat from the
chips. ....
John Levine <johnl@taugh.com> schrieb:
According to c186282 <c186282@nnada.net>:
Umm ... NOT so unusual for the times :-)
Gotta wonder. Big powerful cutting-edge corp
with good history suddenly crashes and burns ......
Sheesh, it's not like it's hard to find out what happened. The world of microprocessors
was different from the world of minicomputers, DEC's management never figured that out,
and its board of directors didn't either.
Gordon Bell wrote this afterword for the "DEC is Dead" book which explains it well.
https://gordonbell.azurewebsites.net/CGB%20Files/DEC_Is_Dead_Bell_Appendix_Schein_Book.pdf
Reading "DEC is Dead", one gets the distinct impression that Gordon
Bell walked on water as a computer architect, but with the benefit
of hindsight, this was not the case.
The specifics of the VAX (especially its addressing modes) made
sure that it would not scale very well towards higher peformance.
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem. The reason they >> need very fast very large databases is that updates have to be serialized
and there's no way to make them faster than the underlying system. Same
for airlines allocating seats and any other database reflecting a real world >> application where there's a finite amount of something to be parcelled out. >>
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone >else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with
me, it can be done in parallel.
The age of the "mainframe" may be coming to an
end alas. 'Cluster'/AI solutions are going to
replace them for most uses. However science/govt
still have uses for big ultra-powerful boxes
for awhile anyway. SOME advantage in keeping
access 'local' too - as opposed to spread all
over the highly-vulnerable net.
On 2026-11-03 04:29, c186282 wrote:
[...]
-a-a The age of the "mainframe" may be coming to an
-a-a end alas. 'Cluster'/AI solutions are going to
-a-a replace them for most uses. However science/govt
-a-a still have uses for big ultra-powerful boxes
-a-a for awhile anyway.
In practice I see no evidence for that here. And why
would it for "science" be better to have a mainframe?
My children actually work "in science" where extremely
demanding complex (hydrodynamic) simulation models are
done; all done on such high-performance clusters. And
this appears to be nowadays quite common as observable.
-a-a SOME advantage in keeping
-a-a access 'local' too - as opposed to spread all
-a-a over the highly-vulnerable net.
There had always been non-local access, even back in
the mainframe days.
There's a more general problem here; whether a single
"box" or a "cluster", if it's connected to the Net it
has to be solidified.
On 11/2/2026 19:53, c186282 wrote:
On 11/2/26 21:49, Lawrence DrCOOliveiro wrote:
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous >>> controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
-a Yep ! Even the old mainframes were "smart" - lots
-a of optimizations that only later migrated down.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
-a System interrupts PER-KEYSTROKE ??? Horrible !
-a The "minis" were pretty good all in all, had a
-a definite 'place'. Lots made. Had some experience
-a with the TI-990 minis. Super-funky hardware-
-a based multi-user scheme with RAM-based registers.
-a Hey, good at the time !
-a I guess you COULD call the PDP-11 a "mini", but
-a imho it didn't have quite the "feel". More like
-a a cut-down mainframe. Good peripheral bus though,
-a you could plug in all kinds of goodies.
Oh, most PDP-11's were definitely minis. Consider that the address space
was 16 bits, i.e. 64 KB. On the 11/45 you could set the MMU to separate instruction fetches into a separate 64KB space from the data
read/writes, but I never got comfortable with that.
The 11/70 could do multi-user with a larger physical memory, so it could
run V6 unix (or the 16-bit BSD version). But nobody should confuse it
with a mainframe.
The genius of the mainframe version of transaction processing was that
the queue of incoming transactions did not have too live in CPU memory
... it sat in the memory of the terminals, waiting for a poll that would bring it inside. This was in complete contrast to the mini computer terminals that were ASYNC TTY, requiring the CPU to catch the character
when the UART interrupted, and the host mini would then decide whether
the character should be echoed back or not.
On 11/3/26 22:47, John Levine wrote:
According to Pancho-a <Pancho.Jones@protonmail.com>:
I know at least one guy working on IBM mainframes, and he tells me >>>>>> they are in high need of good assembler programmers.
They need high i/o with very large databases.
...
Traditional banking transactions almost certainly could be parallelised, >>> to quite a high degree, whether they were or not, I don't know. They are >>> nowadays.
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem.-a The reason
they
need very fast very large databases is that updates have to be serialized
and there's no way to make them faster than the underlying system.-a Same
for airlines allocating seats and any other database reflecting a real
world
application where there's a finite amount of something to be parcelled
out.
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with
me, it can be done in parallel.
When I looked at mainframe technical data IBM put there some
impressive numbers. But I think that is is rather rare that IBM
mainframes are best fit for the problem.
Mainframe may have some advantage in memory bandwidth ...
And why customers use mainframes? That is mostly legacy and risk
av[o]idance.
On Tue, 3 Nov 2026 22:51:30 +0000, David Wade wrote:IBM *tends* to Do It Right.
I think all this shows is that IBM is able to force expensive
hardware refresh by ending support for old machines and old versions
of software so if you need Payment Card Industry compliance you need
supported software.
I donrCOt think thatrCOs true. I have hooked up various clients to various payment systems over the years, and there was never a need for any
particular piece of proprietary software at our end.
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with
me, it can be done in parallel.
NOTICED - my principal bank - a LOT of transactions
come up with a "pending" label. Might take DAYS. This
is not some 3rd-tier bank either.
According to Scott Lurndal <slp53@pacbell.net>:
So, as usual, the facts are not correctly represented by the resident
armchair experts
Dig into a bit and you'll find that those are refreshes for existing
customers, not new customers.
IBM has two markets for z mainframes, the legacy S/360/370/390 market
and high performance linux servers which they call LinuxONE.
I agree they're not likely to have new legacy customers but I'd be surprised that they put all that effort into LinuxONE without expecting to get new customers out of it.
They just announced that the next generation of z will have a
partnership with ARM so the processors can run ARM64 code as well as
the three legacy 360 modes. That tells me they're trying to get
customers who already have the ARM code running in the cloud at AWS or
Azure.
AFAIK several minis had multiplexer channels. It is likely that one
could handle several terminals using single multiplexer channel and drasticaly cut number of interrupts that way. Also I remember
reading about special purpose communiation units that handled
several terminals and put incoming characters in memory.
According to c186282 <c186282@nnada.net>:
I guess you COULD call the PDP-11 a "mini", but
imho it didn't have quite the "feel". More like
a cut-down mainframe. Good peripheral bus though,
you could plug in all kinds of goodies.
I can assure you that PDP-11s were minicomputers. I spent quite a while programming them back in the 1970s.
Oh, most PDP-11's were definitely minis. Consider that the address space
was 16 bits, i.e. 64 KB. On the 11/45 you could set the MMU to separate
instruction fetches into a separate 64KB space from the data
read/writes, but I never got comfortable with that. ...
AFAIK several minis had multiplexer channels. It is likely that
one could handle several terminals using single multiplexer
channel and drasticaly cut number of interrupts that way. Also
I remember reading about special purpose communiation units that
handled several terminals and put incoming characters in memory.
It seems that interrupt per character became popular with full
screen editors: deciding what needs to be handled by software
and what could be handled in the terminal became more complex.
I can assure you that PDP-11s were minicomputers. I spent quite a while
programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
DEC had the VAX line of mainframes ...
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
The PDP-8 I used had "three cycle data break" which
did data transfers into or out of memory (that was
one cycle) using address and word counts also in
memory (those were the other two cycles.) But that
was only for the disks, and didn't do all of the other
chaining and control stuff that channels did.
DEC had the VAX line of mainframes ...
A VAX was a super-mini, not a mainframe. I was there.
Again, simpler I/O even though it was a lot bigger than
a PDP-8 or PDP-11.
If you look through the comp.arch archives you'll find arguments
about whether a PDP-10 was a mainframe or a very large mini. I
claim it was a mini because its I/O was similar to DEC's other
minis, including a hack reminiscent of three-cycle data break.
Others say it was a mainframe because it was marketed that way.
I will admit that the later KL with a PDP-11 front end and MASSBUS
disk I/O was a lot more like a mainframe than the original KA-10.
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
Mainframes were record-oriented, while minis were byte-oriented.
According to Pancho <Pancho.Jones@protonmail.com>:
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem. The reason they >>> need very fast very large databases is that updates have to be serialized >>> and there's no way to make them faster than the underlying system. Same >>> for airlines allocating seats and any other database reflecting a real world
application where there's a finite amount of something to be parcelled out. >>>
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with
me, it can be done in parallel.
Sure, but you're not going to put every account on a different disk. Real transactions are more complicated than that -- when they take the $75 out of my
account they have to credit it somewhere else, and that debit/credit pair has to
be atomic, both work or both fail.
Banks are conservative but they are not stupid. There's a lot of stuff they can
push out onto cheap micro based servers, but there's a certain amount of key business logic that Absolutely Has To Work.
According to c186282 <c186282@nnada.net>:
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with >>> me, it can be done in parallel.
NOTICED - my principal bank - a LOT of transactions
come up with a "pending" label. Might take DAYS. This
is not some 3rd-tier bank either.
That has nothing to do with your bank's database and everything to do
with the way that credit card charges work. First the charge is
approved for an approximate amount which is pretty quick, then the
actual charge is posted which can take anywhere from moments to days.
The amounts are often different, e.g., when you pump gas it approves
an amount big enough to cover the largest plausible fill-up, then when
you're done it posts the final amount. Or at a restaurant, it approves
the amount on the check, then posts that plus whatever tip you add.
For historic reasons it's common to wait to post all of a day's
charges at the end of the day, or even later which is why they can sit
there pending for a while.
According to Pancho <Pancho.Jones@protonmail.com>:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
...
Traditional banking transactions almost certainly could be parallelised,
to quite a high degree, whether they were or not, I don't know. They are
nowadays.
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one succeeded and the other failed*, your bank has a problem. The reason they need very fast very large databases is that updates have to be serialized
and there's no way to make them faster than the underlying system. Same
for airlines allocating seats and any other database reflecting a real world application where there's a finite amount of something to be parcelled out.
* - yes I know about overdrafts, this is an example--
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should
meanwhile be at "best" a niche, and systems programming
with "C" (or similar).
But, yes. It was now for long a problem getting staff for
legacy things. Those companies who are too inert to adapt
[to the technical evolution] will pay, suffer, or demise.
Janis
[...]
On 2026-11-03, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 03 Nov 2026 07:15:04 GMT, Charlie Gibbs wrote:
On 2026-11-03, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 03 Nov 2026 02:53:45 GMT, Charlie Gibbs wrote:
Yes, there is a dichotomy there. Mainframe protocols were designed
for sending files across a continent, while mini protocols were
designed for sending characters across the room. Compromises were
made on both sides.
ThatrCOs not the dichotomy. The dichotomy was that mainframes were
optimized for throughput at the expense of latency, while
minicomputers concentrated on minimizing latency to maximize
interactive responsiveness at the expense of throughput.
Pick your metaphor. In the end the trade-offs are the same.
Nothing rCLmetaphoricalrCY about it. Sending data across a continent in
real-time became commonplace with the coming of ARPAnet. But those
mainframes couldnrCOt actually connect directly, could they? They had to
go through one of those rCLIMPrCY boxes from BBN. And those were built
around ... minicomputers.
That's not the only way to get data across the continent. In
the late '70s our mainframe in Vancouver was driving block-mode
multidropped terminals in Toronto through a leased line at a
blazing 2400 bps. From a software standpoint they looked just
like our local terminals, except that the wires were much longer
and the speed slower. (Our local terminals ran at 9600 bps.)
In those days modems cost about a dollar per bps - that's right,
$2500 for a Bell 202-style modem running synchronously at 2400 bps.
You could get 9600-bps modems, but they were ten grand a pop.
And the leased line wasn't cheap either. As ever, we balanced
price against performance.
On Tue, 3 Nov 2026 22:44:10 +0000, David Wade wrote:
On 03/11/2026 21:57, Lawrence DrCOOliveiro wrote:
no, but how often do you use that?
Neither one had an incremental-search function like the one in
Emacs though, did they? That updates the display in real-time,
after every keystroke, highlighting all the visible matches for the
string as yourCOve typed it so far.
Every day.
So, as usual, the facts are not correctly represented by the residentDig into a bit and you'll find that those are refreshes for existing customers, not new customers.
armchair experts
So, as usual, the facts are not correctly represented by the resident
armchair experts
Dig into a bit and you'll find that those are refreshes for existing
customers, not new customers.
I don't expect new banks, either.
for one type of technology, it was...You could get 9600-bps modems, but they were ten grand a pop.
And the leased line wasn't cheap either.-a As ever, we balanced
price against performance.
And we were told that 9600 was the limit, the theory said so. I think it
was 9600.
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should
meanwhile be at "best" a niche, and systems programming
with "C" (or similar).
But, yes. It was now for long a problem getting staff for
legacy things. Those companies who are too inert to adapt
[to the technical evolution] will pay, suffer, or demise.
Janis
[...]
On Wed, 04 Nov 2026 06:21:03 GMT, Charlie Gibbs wrote:
Mainframes were record-oriented, while minis were byte-oriented.
Maybe better to say rCLminis were character-orientedrCY. Remember DEC had 18-bit and 12-bit minicomputer architectures before IBM even invented
the term rCLbyterCY.
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work. The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
Minis (and later superminis, like the VAX) had it all over mainframes
in that regard.
On 03/11/2026 15:13, Scott Lurndal wrote:
Actually, the numbers decline every year.-a-a The vast majority are
legacy installations (particularly for the Burroughs/Univac/Unisys
mainframes) and there are very few, if any, new customers adopting
mainframes of any flavor.
"Global mainframe market size and IBM Z mainframe revenue trends
highlight a cyclical pattern dominated by multi-year refresh waves (such
as the z16 and z17 cycles):
Global Mainframe Market Size:
-a-a-a-a-a-a-a Valued at approximately $5.33 billion to $5.71 billion in 2025.
-a-a-a-a-a-a-a Projected to grow to $5.65 billion to $6.11 billion in 2026.
-a-a-a-a-a-a-a Forecasted to reach $7.54 billion to $10.53 billion by 2031rCo2034, driven by a compound annual growth rate (CAGR) of roughly
5.95% to 7.05%.
IBM Z Mainframe Performance & Fluctuations:
-a-a-a-a-a-a-a Late 2025 (z17 Launch / Q4): IBM reported its strongest Q4 mainframe revenue in 20 years, with IBM Z sales surging 67%
year-over-year as enterprises adopted the z17 architecture.
-a-a-a-a-a-a-a Mid-2026 (Q2): IBM Z mainframe revenue experienced a short-term
42% year-over-year decline in fiscal Q2 2026 due to delayed large
capital expenditure (capex) deals and client shifts toward
supply-constrained servers, though management noted these were deferred purchases rather than lost contracts (with roughly one-third closing
shortly after the quarter)."
So, as usual, the facts are not correctly represented by the resident armchair experts
On Tue, 3 Nov 2026 12:45:09 +0100, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
Something not quite right there. Hand-coded assembly language only
makes sense for minimizing CPU usage. If yourCOre doing lots of I/O,
then the CPU is not the bottleneck, so trying to hand-tune assembly
code is a waste of time.
On 03/11/2026 20:25, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe
thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work. The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
Minis (and later superminis, like the VAX) had it all over mainframes
in that regard.
Why? I don't need to imagine, I did it all the time. Its not a problem.
The editor generates a screen full of data.
The controller handles most of the character editing so you change the
local buffer which does see every key stroke.
At some point you send the modified screen back to the host.
Personally I prefer the XEDIT the VM/CMS editor to the VMS one, but then
I am more familiar with it.
Dave
On Tue, 3 Nov 2026 18:06:33 +0100, Carlos E. R. wrote:
And extra cost by ensuring it is reliable.
Very expensive way of ensuring reliability. Google ensures reliability
by running (as of a count in a paper published a decade or two ago)
about half a million servers. So, at any moment, a dozen or a hundred
of them are down; so what? The customers donrCOt notice.
Neither one had an incremental-search function like the one in Emacs
though, did they? That updates the display in real-time, after every keystroke, highlighting all the visible matches for the string as
yourCOve typed it so far.
Now try doing it with a regular-expression search ...
If you want heavy usage, I would guess bank risk calcs, weather calcs,
lots of science applications, running on compute grids, (or whatever the modern word is), blow this mainframe high i/o away.
Lars Poulsen <lars@beagle-ears.com> wrote:
But the RISC instruction sets could be
done in a much simpler hardware package, allowing for smaller and
cheaper chips, and the compilers became much better (because if the
instruction set is a nightmare, everything needs to be compiled).
I disagree with "instruction set is a nightmare".
According to Scott Lurndal <slp53@pacbell.net>:
So, as usual, the facts are not correctly represented by the resident >>>armchair experts
Dig into a bit and you'll find that those are refreshes for existing >>customers, not new customers.
IBM has two markets for z mainframes, the legacy S/360/370/390 market
and high performance linux servers which they call LinuxONE.
I agree they're not likely to have new legacy customers but I'd be surprised >that they put all that effort into LinuxONE without expecting to get new >customers out of it.
They just announced that the next generation of z will have a
partnership with ARM so the processors can run ARM64 code as well as
the three legacy 360 modes. That tells me they're trying to get
customers who already have the ARM code running in the cloud at AWS or
Azure.
On Tue, 3 Nov 2026 20:53:29 +0000, David Wade wrote:
On 03/11/2026 20:25, Lawrence DrCOOliveiro wrote:
Imagine trying to run a full-screen editor that way: it just
wouldnrCOt work. The app really needed to be able to respond to every
keystroke in order to maintain the necessary level of interactive
response. Minis (and later superminis, like the VAX) had it all
over mainframes in that regard.
Why? I don't need to imagine, I did it all the time. Its not a
problem. The editor generates a screen full of data. The controller
handles most of the character editing so you change the local buffer
which does see every key stroke. At some point you send the modified
screen back to the host.
Personally I prefer the XEDIT the VM/CMS editor to the VMS one, but
then I am more familiar with it.
Neither one had an incremental-search function like the one in Emacs
though, did they?
On Tue, 3 Nov 2026 05:28:52 -0800, Lars Poulsen wrote:
By the time of VMS, there was a new paradigm, where you could push
the pre-processing out to the terminal cluster: The TTY device in
the OS had mode bitmasks, specifying for each character whether it
should be echoed and whether it should generate a read completion
interrupt.
You realize, all this was done in *software* in the terminal driver,
not in the terminal controller?
So yes, the CPU was still getting an interrupt on every incoming
character.
Unix got a similar set of modes controlled with the stty command and
its associated ioctl system calls, but I don't think I have ever
seen these succesfully be used to put the TTY pre-processing in a
remote front-end controller, like VMS could.
Unix never had a terminal driver as advanced as VMS did, that is true.
On Wed, 4 Nov 2026 02:07:00 -0000 (UTC), Waldek Hebisch wrote:
When I looked at mainframe technical data IBM put there some
impressive numbers. But I think that is is rather rare that IBM
mainframes are best fit for the problem.
IBM has been coasting on inertia since the 1980s.
Consider: in 1980, there were something like 5 (?) different computer
companies in the mainframe business. Two decades later, that had dropped
down to 1.
According to Pancho <Pancho.Jones@protonmail.com>:
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem. The reason they >>> need very fast very large databases is that updates have to be serialized >>> and there's no way to make them faster than the underlying system. Same >>> for airlines allocating seats and any other database reflecting a real world
application where there's a finite amount of something to be parcelled out. >>>
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with
me, it can be done in parallel.
Sure, but you're not going to put every account on a different disk. Real transactions are more complicated than that -- when they take the $75 out of my
account they have to credit it somewhere else, and that debit/credit pair has to
be atomic, both work or both fail.
Banks are conservative but they are not stupid. There's a lot of stuff they can
push out onto cheap micro based servers, but there's a certain amount of key business logic that Absolutely Has To Work.
On 11/3/26 18:18, Pancho wrote:
On 11/3/26 22:47, John Levine wrote:
According to Pancho-a <Pancho.Jones@protonmail.com>:
I know at least one guy working on IBM mainframes, and he tells me >>>>>>> they are in high need of good assembler programmers.
They need high i/o with very large databases.
...
Traditional banking transactions almost certainly could be
parallelised,
to quite a high degree, whether they were or not, I don't know. They
are
nowadays.
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem.-a The reason
they
need very fast very large databases is that updates have to be
serialized
and there's no way to make them faster than the underlying system.-a Same >>> for airlines allocating seats and any other database reflecting a
real world
application where there's a finite amount of something to be
parcelled out.
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do
with me, it can be done in parallel.
-a NOTICED - my principal bank - a LOT of transactions
-a come up with a "pending" label. Might take DAYS. This
-a is not some 3rd-tier bank either.
-a Two or more larger credit/debit card charges, not
-a sure HOW they 'compete' in that system. If too
-a near each other one MIGHT get bumped. Problem !
-a In short, even 'online/instant' banking may not
-a be THAT instant.
On 11/3/26 13:27, The Natural Philosopher wrote:
On 03/11/2026 15:13, Scott Lurndal wrote:
Actually, the numbers decline every year.-a-a The vast majority are
legacy installations (particularly for the Burroughs/Univac/Unisys
mainframes) and there are very few, if any, new customers adopting
mainframes of any flavor.
"Global mainframe market size and IBM Z mainframe revenue trends
highlight a cyclical pattern dominated by multi-year refresh waves (such
as the z16 and z17 cycles):
Global Mainframe Market Size:
-a-a-a-a-a-a-a-a Valued at approximately $5.33 billion to $5.71 billion in 2025.
-a-a-a-a-a-a-a-a Projected to grow to $5.65 billion to $6.11 billion in 2026.
-a-a-a-a-a-a-a-a Forecasted to reach $7.54 billion to $10.53 billion by
2031rCo2034, driven by a compound annual growth rate (CAGR) of roughly
5.95% to 7.05%.
IBM Z Mainframe Performance & Fluctuations:
-a-a-a-a-a-a-a-a Late 2025 (z17 Launch / Q4): IBM reported its strongest Q4 >> mainframe revenue in 20 years, with IBM Z sales surging 67%
year-over-year as enterprises adopted the z17 architecture.
-a-a-a-a-a-a-a-a Mid-2026 (Q2): IBM Z mainframe revenue experienced a short-term
42% year-over-year decline in fiscal Q2 2026 due to delayed large
capital expenditure (capex) deals and client shifts toward
supply-constrained servers, though management noted these were deferred
purchases rather than lost contracts (with roughly one-third closing
shortly after the quarter)."
So, as usual, the facts are not correctly represented by the resident
armchair experts
My only question is how many mainframe shipments represent NEW
customers, as opposed to additional horsepower for current customers.
I suppose, given that the trend today is to consolidation of everything, that eventually everything will be owned by BigCorp, who will then be
the only mainframe customer.
On 2026-11-04 00:05, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 22:44:10 +0000, David Wade wrote:
On 03/11/2026 21:57, Lawrence DrCOOliveiro wrote:
no, but how often do you use that?
Neither one had an incremental-search function like the one in
Emacs though, did they? That updates the display in real-time,
after every keystroke, highlighting all the visible matches for the
string as yourCOve typed it so far.
Every day.
I'm told that most people do not work directly on a terminal on the mainframe. There is not an OS proper...
On 11/3/26 13:25, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe
thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work. The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
Minis (and later superminis, like the VAX) had it all over mainframes
in that regard.
I violently disagree. Full screen editors provide instant response to
most operations: Move the cursor around, add, delete, and change text. There's only a delay when you're done editing that screen and want to
move on to something else. On a well-tuned system this is a slight delay.
It just occurred to me that I'm typing this text into a mail app which
is basically the same thing. I can edit all I want, and then press SEND
when I'm done.
Maybe it's my mainframe background, but I think having "the computer" respond instantly to every keystroke is just stupid.
Peter Flass wrote this screed in ALL-CAPS (fixed):[...]
It just occurred to me that I'm typing this text into a mail app which
is basically the same thing. I can edit all I want, and then press SEND
when I'm done.
Maybe it's my mainframe background, but I think having "the computer"
respond instantly to every keystroke is just stupid.
I demand instant response!
On 11/3/26 21:16, Janis Papanagnou wrote:[...]
My children actually work "in science" where extremely
demanding complex (hydrodynamic) simulation models are
done; all done on such high-performance clusters. And
this appears to be nowadays quite common as observable.
-a Umm ... your KIDS are into hydrodynamics ? Are
-a they 'kids' anymore ?
There had always been non-local access, even back in
the mainframe days.
-a Yep - happened. However it's easier to guard/armor
-a ONE mainframe than a vast network.
There's a more general problem here; whether a single
"box" or a "cluster", if it's connected to the Net it
has to be solidified.
-a DOES require a lot of 'solidification' - esp now.
-a Otherwise you're just working for your worst enemies.
On 11/3/26 23:15, John Levine wrote:
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
Kinda true.
However 'minis' were SUPPOSED to do everything
mainframes could do - just a a DISCOUNT PRICE.
A VAX was a super-mini, not a mainframe. I was there.
Again, simpler I/O even though it was a lot bigger than
a PDP-8 or PDP-11.
Ummmmm ... I suspect we're dealing with
slightly different defs of 'mainframe'
and 'mini' and whatever. There is NO
clear line.
On 11/4/26 02:10, Lawrence DrCOOliveiro wrote:
On Wed, 04 Nov 2026 06:21:03 GMT, Charlie Gibbs wrote:
Mainframes were record-oriented, while minis were byte-oriented.
Maybe better to say rCLminis were character-orientedrCY. Remember DEC had
18-bit and 12-bit minicomputer architectures before IBM even invented
the term rCLbyterCY.
Well, the 'standard' 8-bit byte kind of evolved, or converged.
On 04/11/2026 02:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:You don't know what IBM assembler is, do you?
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should
meanwhile be at "best" a niche, and systems programming
with "C" (or similar).
But, yes. It was now for long a problem getting staff forYou don't really know what you are talking about, do you?
legacy things. Those companies who are too inert to adapt
[to the technical evolution] will pay, suffer, or demise.
On 2026-11-04 12:25, The Natural Philosopher wrote:
On 04/11/2026 02:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:You don't know what IBM assembler is, do you?
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should
meanwhile be at "best" a niche, and systems programming
with "C" (or similar).
"IBM assembler" specifically? No. - I know (and programmed in) a
couple other "high-level" macro-assemblers. - Do you imply that
the "IBM assembler" is somehow special, and so different that it
would somehow invalidate what I said?
But, yes. It was now for long a problem getting staff forYou don't really know what you are talking about, do you?
legacy things. Those companies who are too inert to adapt
[to the technical evolution] will pay, suffer, or demise.
I certainly do. - But, given your post, I doubt you intend to
contribute any substance and prefer just playing the troll.
(With your stance I don't think it's worthwhile to respond to
you in future.) Good luck.
Janis
DOOM and DUKE NUKEM were great. QUAKE ... it
was mostly BLACK SCREEN shit, why bother having
a monitor ???
Still have DUKE stashed away somewhere, runs in
DOSBox. Lots of user-created episodes too. For one
I had to write a 256 bit binary-2-decimal calculator :-)
Peter Flass wrote this screed in ALL-CAPS (fixed):
Maybe it's my mainframe background, but I think having "the computer"
respond instantly to every keystroke is just stupid.
I demand instant response!
And if you think the fruit vendor is better, you can rest easy, because
at one time they refused to update my card number. Actually refused.
After reaching the /right/ support person, that person could not help
figure out why the website told me the card number had been updated, but
then the TV application always tried to use the old card number.
On 11/3/26 13:29, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 12:45:09 +0100, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
Something not quite right there. Hand-coded assembly language only
makes sense for minimizing CPU usage. If yourCOre doing lots of I/O,
then the CPU is not the bottleneck, so trying to hand-tune assembly
code is a waste of time.
I kind of agree. As a former assembler programmer - at one time it was
my first language - I'd like it to be true (although I'm out of the job market), but I can't see it. These days most HLL code is more optimized than a lot of assembler. A good assembler programmer would not make a
lot of these optimizations due to lack of readability.
On 04/11/2026 09:58, Carlos E. R. wrote:
You could get 9600-bps modems, but they were ten grand a pop.
And the leased line wasn't cheap either.-a As ever, we balanced
price against performance.
And we were told that 9600 was the limit, the theory said so. I think it
was 9600.
for one type of technology, it was...
Lawrence =?iso-8859-13?q?D=FFOliveiro?= <ldo@nz.invalid> writes:
On Wed, 4 Nov 2026 02:07:00 -0000 (UTC), Waldek Hebisch wrote:
When I looked at mainframe technical data IBM put there some
impressive numbers. But I think that is is rather rare that IBM
mainframes are best fit for the problem.
IBM has been coasting on inertia since the 1980s.
Consider: in 1980, there were something like 5 (?) different computer
There were a BUNCH of different mainframe companies.
On 11/3/26 13:53, David Wade wrote:
Personally I prefer the XEDIT the VM/CMS editor to the VMS one, but then
I am more familiar with it.
I also like XEdit, in fact I run THE on my PC, although it is not my
primary editor. VAX TPU is also quite good, however, but we had great response time on our VAX.
On 11/3/26 15:15, Pancho wrote:
If you want heavy usage, I would guess bank risk calcs, weather calcs,
lots of science applications, running on compute grids, (or whatever
the modern word is), blow this mainframe high i/o away.
I think back in the day the quants preferred Sun.
On 2026-11-04, Johann 'Myrkraverk' Oskarsson <johann@myrkraverk.invalid> wrote:
And if you think the fruit vendor is better, you can rest easy, because
at one time they refused to update my card number. Actually refused.
After reaching the /right/ support person, that person could not help
figure out why the website told me the card number had been updated, but
then the TV application always tried to use the old card number.
Ah, you've been around the mulberry bush with Netflix too, I presume.
We recently found we couldn't log in to Netflix because they didn't
like our card number - and quoted the last four digits and date of
a card that expired seven years ago. I wonder how they had managed
to bill us the previous month.
I called their customer support line, and the droid became downright belligerent, claiming that it wasn't their fault but our bank's.
I wound up hanging up on him, and we cancelled our Netflix account.
On 2026-11-04, Chris Ahlstrom <OFeem1987@teleworm.us> wrote:
Peter Flass wrote this screed in ALL-CAPS (fixed):
Maybe it's my mainframe background, but I think having "the computer"
respond instantly to every keystroke is just stupid.
I can understand that. The overhead involved with dealing with every character coming in from multiple terminals can be considerable.
Given the CPU power we had back in the day (maybe 130 KIPS for a small-to-middling mainframe), it would bring the machine to its knees.
I demand instant response!
Given typical mainframe protocols of the day (bisync, half-duplex, multidropped terminals that get polled maybe once a second), you're
not going to get that even if you had all the CPU power in the world.
I remember reading about an experiment done in an IBM shop where
they shoveled in enough hardware and tweaked the protocols so that
terminal users got sub-second response time. Apparently it increased throughput even more than one would expect by just scaling the
response time - users could develop a rhythm and not have their
minds wander while waiting for a response.
On the other side of the coin, someone at Univac decided that
the key to productivity was not so much a quick response time
as a consistent one - which led to their system inserting
delays if a response was ready too soon. :-p
On 11/3/26 15:16, Waldek Hebisch wrote:
I disagree with "instruction set is a nightmare".
Used to be a nightmare. Something like putting delay slots after
branch instructions.
If you look through the comp.arch archives you'll find arguments about whether a PDP-10 was a mainframe or a very large mini. I claim it was a mini because its I/O was similar to DEC's other minis, including a hack reminiscent of three-cycle data break. Others say it was a mainframe because it was marketed that way.
I will admit that the later KL with a PDP-11 front end and MASSBUS disk I/O was a lot more like a mainframe than the original KA-10.
You can't distribute an airline reservation system, the data has to
live in one place.
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
On Wed, 04 Nov 2026 06:21:03 GMT, Charlie Gibbs wrote:
Mainframes were record-oriented, while minis were byte-oriented.
Maybe better to say "minis were character-oriented". Remember DEC had
18-bit and 12-bit minicomputer architectures before IBM even invented
the term "byte".
On 11/3/26 14:57, Lawrence DrCOOliveiro wrote:
Neither one had an incremental-search function like the one in
Emacs though, did they? That updates the display in real-time,
after every keystroke, highlighting all the visible matches for the
string as yourCOve typed it so far.
I find this more distracting than useful.
Let me type my search, including correcting inevitable typos, etc.
and then run it instead of continually flashing the screen at me as
I type.
Now try doing it with a regular-expression search ...
Why would this be a problem? "Yes, ISPF View and Edit support regular expressions."
And we were told that 9600 was the limit, the theory said so. I
think it was 9600.
On 11/3/26 13:25, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe
thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work. The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
Minis (and later superminis, like the VAX) had it all over mainframes
in that regard.
I violently disagree. Full screen editors provide instant response to
most operations: Move the cursor around, add, delete, and change text. There's only a delay when you're done editing that screen and want to
move on to something else. On a well-tuned system this is a slight delay.
It just occurred to me that I'm typing this text into a mail app which
is basically the same thing. I can edit all I want, and then press SEND
when I'm done.
Maybe it's my mainframe background, but I think having "the computer" respond instantly to every keystroke is just stupid.
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written
in C, with very little assembly language.
According to Pancho <Pancho.Jones@protonmail.com>:
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem. The reason they >>> need very fast very large databases is that updates have to be serialized >>> and there's no way to make them faster than the underlying system. Same >>> for airlines allocating seats and any other database reflecting a real world
application where there's a finite amount of something to be parcelled out. >>>
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone >>else's account need to be on the same database/compute server as my >>account? If A and B are doing a transaction, that has nothing to do with >>me, it can be done in parallel.
Sure, but you're not going to put every account on a different disk. Real transactions are more complicated than that -- when they take the $75 out of my
account they have to credit it somewhere else, and that debit/credit pair has to
be atomic, both work or both fail.
Banks are conservative but they are not stupid. There's a lot of stuff they can
push out onto cheap micro based servers, but there's a certain amount of key business logic that Absolutely Has To Work.
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written
in C, with very little assembly language.
On 2026-11-04, The Natural Philosopher <tnp@invalid.invalid> wrote:
On 04/11/2026 09:58, Carlos E. R. wrote:
You could get 9600-bps modems, but they were ten grand a pop.
And the leased line wasn't cheap either.-a As ever, we balanced
price against performance.
And we were told that 9600 was the limit, the theory said so. I think it >>> was 9600.
for one type of technology, it was...
The mainframe I worked on could do async (although nobody
ever did) - but since Everybody Knows that async is a
low-speed protocol the ports weren't designed to run async
any faster than 2400 bps, even for direct connections.
Didn't part of the RS-232 spec say that wiring had to handle
50-foot runs at 9600 bps? Given cable capacitance it could
be a crap shoot, but we managed to get async terminals
(or emulators on PCs) to talk to a Sun workstation at the
other end of the building at 38400 bps.
Also, the disks do not have a filesystem, either (I also asked).
Not something I ever did, but I strongly suspect it is message
queues, including acknowledgements. Tricky to get perfect, but
standard IT stuff.
I got my PC to talk to another PC at 118K with a serial port across
perhaps 20 metres, in the 90's.
On 2026-11-04, Peter Flass <Peter@Iron-Spring.com> wrote:
On 11/3/26 13:29, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 12:45:09 +0100, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
Something not quite right there. Hand-coded assembly language only
makes sense for minimizing CPU usage. If yourCOre doing lots of I/O,
then the CPU is not the bottleneck, so trying to hand-tune assembly
code is a waste of time.
I kind of agree. As a former assembler programmer - at one time it was
my first language - I'd like it to be true (although I'm out of the job
market), but I can't see it. These days most HLL code is more optimized
than a lot of assembler. A good assembler programmer would not make a
lot of these optimizations due to lack of readability.
A mediocre assembly language programmer would not let lack of
readability stand in the way. And once you get into self-modifying
code there are all sorts of horrible things you can do.
On 11/4/26 14:54, Peter Flass wrote:
On 11/3/26 15:15, Pancho wrote:
If you want heavy usage, I would guess bank risk calcs, weather
calcs, lots of science applications, running on compute grids, (or
whatever the modern word is), blow this mainframe high i/o away.
I think back in the day the quants preferred Sun.
Not really,-quants used Excel on PCs. Lotus123 ran on Sun, but it was
second rate. I think prior to Windows NT some IT departments gambled on
Sun, which had a proper OS, rather than PCs, but it quickly became
obvious it was a poor choice.
There was also a buzz about RISC cpus replacing x86, which to be fair
now seems to be happening, just 35 years late.
In alt.folklore.computers John Levine <johnl@taugh.com> wrote:
According to Pancho <Pancho.Jones@protonmail.com>:
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one >>>> succeeded and the other failed*, your bank has a problem. The reason they >>>> need very fast very large databases is that updates have to be serialized >>>> and there's no way to make them faster than the underlying system. Same >>>> for airlines allocating seats and any other database reflecting a real world
application where there's a finite amount of something to be parcelled out.
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with >>> me, it can be done in parallel.
Sure, but you're not going to put every account on a different disk. Real
transactions are more complicated than that -- when they take the $75 out of my
account they have to credit it somewhere else, and that debit/credit pair has to
be atomic, both work or both fail.
This is somewhat different for credit cards. I have a debit card and
when I look at my account I see that some payment are done while
other show "blocked" status. AFAIK "blocked" means that transation
was authorised, but there was no money transfer. All what needs
to be done in real time is authorisation, once transation is
authorised merchant is assured of payment so can treat is a
finished. But actual transfer of money happens with some delay
and could be done by high latency high bandwidth process
(say "nightly batch window").
The only thing which must happen in real time is updating
status of my account and confirmation of autorisation (possibly
with some magic tokens) to the merchant, which involves only
merchant system and machine handling my account. So my bank
could split accounts between several machines (or rather sets
of machines as data almost surely is replicated to ensure
reliablity).
BTW: AFAIK all interbank transfers by law much go trough
national bank and this introduces some delay. IIUC in the
past this used to be something like hours or a day, now
it seem to be down to minutes. Way too slow for credit
card transaction, so that may be one of reasons for
splitting them into authorisation part and actual transfer.
Banks are conservative but they are not stupid. There's a lot of stuff they can
push out onto cheap micro based servers, but there's a certain amount of key >> business logic that Absolutely Has To Work.
Machines that handle critcal things in banks may be
significantly cheaper than IBM mainframes, but they are not
exactly cheap. Those are server class machines choosen for
high reliability and high performance. I would be not
surprised if they used similar machines even for not
critical task, as having single/few kinds of server machines
simplifies logistic.
BTW2: I have read that banking system of Ukraine is now in
the cloud outside of Ukraine (and similar for many government
services). I have not heard of mainframe cloud offerings
from major cloud vendors (IIRC Ukraine contracted service
from one of major cloud vendors) so it is quite likely that
they use no mainframes.
x86 is RISC under the hood.
On Wed, 4 Nov 2026 19:04:22 -0700, Peter Flass wrote:
x86 is RISC under the hood.
ThatrCOs nonsense. ThatrCOs Intel/AMD PR talking.
On 11/4/26 03:12, John Levine wrote:
According to Pancho <Pancho.Jones@protonmail.com>:
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one >>>> succeeded and the other failed*, your bank has a problem. The reason they >>>> need very fast very large databases is that updates have to be serialized >>>> and there's no way to make them faster than the underlying system. Same >>>> for airlines allocating seats and any other database reflecting a real world
application where there's a finite amount of something to be parcelled out.
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with >>> me, it can be done in parallel.
Sure, but you're not going to put every account on a different disk. Real
transactions are more complicated than that -- when they take the $75 out of my
account they have to credit it somewhere else, and that debit/credit pair has to
be atomic, both work or both fail.
Not something I ever did, but I strongly suspect it is message queues, including acknowledgements. Tricky to get perfect, but standard IT
stuff.
Banks are conservative but they are not stupid. There's a lot of stuff they can
push out onto cheap micro based servers, but there's a certain amount of key >> business logic that Absolutely Has To Work.
Absolutely has to work, means testing. That is why they keep the old
stuff, it has been tested, in production.
Newer system reliability will be almost entirely about software
design, better techniques, but testing is difficult.
We recently had a scandal in the UK, the Post Office, shocking flaws
in design, from the mid 1990s, but over the years they caught most of
the bugs and made it work.
A superficial look at the news reports--
showed they weren't using things like idempotent operations. Modern
systems are designed to be much more tolerant of hardware glitches.
According to c186282 <c186282@nnada.net>:
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with >>> me, it can be done in parallel.
NOTICED - my principal bank - a LOT of transactions
come up with a "pending" label. Might take DAYS. This
is not some 3rd-tier bank either.
That has nothing to do with your bank's database and everything to do
with the way that credit card charges work. First the charge is
approved for an approximate amount which is pretty quick, then the
actual charge is posted which can take anywhere from moments to days.
The amounts are often different, e.g., when you pump gas it approves
an amount big enough to cover the largest plausible fill-up, then when
you're done it posts the final amount. Or at a restaurant, it approves
the amount on the check, then posts that plus whatever tip you add.
For historic reasons it's common to wait to post all of a day's--
charges at the end of the day, or even later which is why they can sit
there pending for a while.
On Wed, 4 Nov 2026 07:52:35 -0700, Peter Flass wrote:
On 11/3/26 14:57, Lawrence DrCOOliveiro wrote:
Neither one had an incremental-search function like the one in
Emacs though, did they? That updates the display in real-time,
after every keystroke, highlighting all the visible matches for the
string as yourCOve typed it so far.
I find this more distracting than useful.
You do, others donrCOt.
Let me type my search, including correcting inevitable typos, etc.
and then run it instead of continually flashing the screen at me as
I type.
Except you may not notice your typos until you see what does or
doesnrCOt match the search. ThatrCOs why the immediate feedback is useful.
--Now try doing it with a regular-expression search ...
Why would this be a problem? "Yes, ISPF View and Edit support regular
expressions."
But they are even more difficult to get right. Which is why the
immediate feedback becomes even more useful.
We recently had a scandal in the UK, the Post Office, shocking flaws
in design, from the mid 1990s, but over the years they caught most of
the bugs and made it work.
At the same time they kept prosecuting innocent people for it, right?
On Thu, 5 Nov 2026 00:14:16 +0100, Carlos E. R. wrote:
Also, the disks do not have a filesystem, either (I also asked).
Programming like itrCOs 1975 <https://vinyl-cache.org/docs/9.1/phk/notes.html>:
So whatrCOs wrong with 1975 programming ?
The really short answer is that computers do not have two kinds of
storage anymore.
...
Virtual memory was meant to make it easier to program when data
was larger than the physical memory, but people have still not
caught on.
On 11/4/26 12:51, Charlie Gibbs wrote:
On 2026-11-04, Peter Flass <Peter@Iron-Spring.com> wrote:
On 11/3/26 13:29, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 12:45:09 +0100, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
Something not quite right there. Hand-coded assembly language only
makes sense for minimizing CPU usage. If yourCOre doing lots of I/O,
then the CPU is not the bottleneck, so trying to hand-tune assembly
code is a waste of time.
I kind of agree. As a former assembler programmer - at one time it was
my first language - I'd like it to be true (although I'm out of the job
market), but I can't see it. These days most HLL code is more optimized >>> than a lot of assembler. A good assembler programmer would not make a
lot of these optimizations due to lack of readability.
A mediocre assembly language programmer would not let lack of
readability stand in the way. And once you get into self-modifying
code there are all sorts of horrible things you can do.
1. "standard" things like converting a branch to a nop or back. I know
1401 Autocoder people used to do horrible things like using part of an >instruction as a constant.
On 2026-11-04, The Natural Philosopher <tnp@invalid.invalid> wrote:
On 04/11/2026 09:58, Carlos E. R. wrote:
You could get 9600-bps modems, but they were ten grand a pop.
And the leased line wasn't cheap either.-a As ever, we balanced
price against performance.
And we were told that 9600 was the limit, the theory said so. I think it >>> was 9600.
for one type of technology, it was...
The mainframe I worked on could do async (although nobody
ever did) - but since Everybody Knows that async is a
low-speed protocol the ports weren't designed to run async
any faster than 2400 bps, even for direct connections.
Didn't part of the RS-232 spec say that wiring had to handle
50-foot runs at 9600 bps? Given cable capacitance it could
be a crap shoot, but we managed to get async terminals
(or emulators on PCs) to talk to a Sun workstation at the
other end of the building at 38400 bps.
On Wed, 4 Nov 2026 07:58:35 -0700, Peter Flass wrote:
On 11/3/26 15:16, Waldek Hebisch wrote:
I disagree with "instruction set is a nightmare".
Used to be a nightmare. Something like putting delay slots after
branch instructions.
There was this brief time when it seemed like the hardware would be
simpler if this was done. That window was just a few years in about
the mid-1980s, just wide enough to catch SPARC and MIPS in the net
(maybe PA-RISC as well?).
In alt.folklore.computers Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
On 2026-11-04, The Natural Philosopher <tnp@invalid.invalid> wrote:
On 04/11/2026 09:58, Carlos E. R. wrote:
You could get 9600-bps modems, but they were ten grand a pop.
And the leased line wasn't cheap either.-a As ever, we balanced
price against performance.
And we were told that 9600 was the limit, the theory said so. I think it >>>> was 9600.
for one type of technology, it was...
The mainframe I worked on could do async (although nobody
ever did) - but since Everybody Knows that async is a
low-speed protocol the ports weren't designed to run async
any faster than 2400 bps, even for direct connections.
Didn't part of the RS-232 spec say that wiring had to handle
50-foot runs at 9600 bps? Given cable capacitance it could
be a crap shoot, but we managed to get async terminals
(or emulators on PCs) to talk to a Sun workstation at the
other end of the building at 38400 bps.
RS-232 is crappy from signaling point of view: too large
voltage and too large resistance which leads to mismatch
with typical tarsmition lines (like in building telephone
wires). IIUC maiframes had special adapters for better
signaling. This was mainly used with synchronous transmission,
but AFAICS the main thing to get better speed was better
electrical part. Synchronous may gain some fraction of
speed because there is no need to transmit start and stop
bits required by async. If clock is transmited by separarte
line, then synchronous hardware may be much simpler. OTOH
at high speed and with long lines separate lines tend to
have different delay, which means that separate clock no
longer is in sync with the signal. I am not sure how this
was resolved around 1970 (maybe distances and speed were low
enough that it did not matter), but now separate clock
is limited to short distances. Once distance and speed
are large enough, then it is async or its variations
(which send less data but still allow clock recovery
in the receiver). It seems that people also used
term "synchronous" for case when there was no separate
clock line. Schemes like MFM are simple, but less
efficient than async, and once one moves to more advanced
schemes it is more like async and unlike simple
synchronous schemes.
On 11/4/26 12:51, Charlie Gibbs wrote:
On 2026-11-04, Peter Flass <Peter@Iron-Spring.com> wrote:
I kind of agree. As a former assembler programmer - at one time it was
my first language - I'd like it to be true (although I'm out of the job
market), but I can't see it. These days most HLL code is more optimized >>> than a lot of assembler. A good assembler programmer would not make a
lot of these optimizations due to lack of readability.
A mediocre assembly language programmer would not let lack of
readability stand in the way. And once you get into self-modifying
code there are all sorts of horrible things you can do.
1. "standard" things like converting a branch to a nop or back.
I know 1401 Autocoder people used to do horrible things like using
part of an instruction as a constant.
2. You can do anything as long as you comment it. I once had a sequence
to test-store-conditional branch, for some reason I don't recall, bit I tagged it with comments saying "these three instructions must be kept together".
On 2026-11-04, John Levine <johnl@taugh.com> wrote:
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
I agree. I've always believed that the difference between
a mainframe and a mini had nothing to do with the size of
the box, but everything to do with architecture, particularly
I/O. Mainframes were record-oriented, while minis were
byte-oriented.
On 11/3/26 13:27, The Natural Philosopher wrote:
On 03/11/2026 15:13, Scott Lurndal wrote:
Actually, the numbers decline every year.-a-a The vast majority are
legacy installations (particularly for the Burroughs/Univac/Unisys
mainframes) and there are very few, if any, new customers adopting
mainframes of any flavor.
"Global mainframe market size and IBM Z mainframe revenue trends
highlight a cyclical pattern dominated by multi-year refresh waves (such
as the z16 and z17 cycles):
Global Mainframe Market Size:
-a-a-a-a-a-a-a-a Valued at approximately $5.33 billion to $5.71 billion in 2025.
-a-a-a-a-a-a-a-a Projected to grow to $5.65 billion to $6.11 billion in 2026.
-a-a-a-a-a-a-a-a Forecasted to reach $7.54 billion to $10.53 billion by
2031rCo2034, driven by a compound annual growth rate (CAGR) of roughly
5.95% to 7.05%.
IBM Z Mainframe Performance & Fluctuations:
-a-a-a-a-a-a-a-a Late 2025 (z17 Launch / Q4): IBM reported its strongest Q4 >> mainframe revenue in 20 years, with IBM Z sales surging 67%
year-over-year as enterprises adopted the z17 architecture.
-a-a-a-a-a-a-a-a Mid-2026 (Q2): IBM Z mainframe revenue experienced a short-term
42% year-over-year decline in fiscal Q2 2026 due to delayed large
capital expenditure (capex) deals and client shifts toward
supply-constrained servers, though management noted these were deferred
purchases rather than lost contracts (with roughly one-third closing
shortly after the quarter)."
So, as usual, the facts are not correctly represented by the resident
armchair experts
My only question is how many mainframe shipments represent NEW
customers, as opposed to additional horsepower for current customers.
I suppose, given that the trend today is to consolidation of everything, that eventually everything will be owned by BigCorp, who will then be
the only mainframe customer.
A mediocre assembly language programmer would not let lack of
readability stand in the way. And once you get into self-modifying
code there are all sorts of horrible things you can do.
On 11/3/26 15:15, Pancho wrote:
If you want heavy usage, I would guess bank risk calcs, weather calcs,
lots of science applications, running on compute grids, (or whatever
the modern word is), blow this mainframe high i/o away.
I think back in the day the quants preferred Sun.
Do I have something to say on this subject?-a No, not really; I just find
it amusing when trolls accuse other people of being trolls, and then, presumably, killfile the person.-a Consider it "trash taking itself out."
Best wishes, and happy philosophering of natural phenomena!
Lawrence =?iso-8859-13?q?D=FFOliveiro?= <ldo@nz.invalid> writes:
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written >> in C, with very little assembly language.
Again with the incorrect characterizations.
The fastest supercomputers are running massive amounts of
modern (and some ancient) Fortran code.
On 2026-11-04 22:47, Lawrence DrCOOliveiro wrote:
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written >> in C, with very little assembly language.
Well, I know for a fact that this machine doesn't run any standard OS. I asked him. Also, the disks do not have a filesystem, either (I also
asked). Of course most of the programmers work in some high level
language, but there are certain critical operations that are optimized
in assembler. I don't remember which, and obviously I'm not going to
quote him breaking confidence. But if anyone is a good programmer in
those machines, in assembler, they are hiring.
On 04/11/2026 23:14, Carlos E. R. wrote:
On 2026-11-04 22:47, Lawrence DrCOOliveiro wrote:My mathematical friend needs to use assembler routines to access some of
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me >>>>>> they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written >>> in C, with very little assembly language.
Well, I know for a fact that this machine doesn't run any standard OS.
I asked him. Also, the disks do not have a filesystem, either (I also
asked). Of course most of the programmers work in some high level
language, but there are certain critical operations that are optimized
in assembler. I don't remember which, and obviously I'm not going to
quote him breaking confidence. But if anyone is a good programmer in
those machines, in assembler, they are hiring.
the vector processing that the chipset, but not available-a C libraries, supports.
IBM Macro-a assembler is practically a high level language, given the
scale of macro usage in legacy systems
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
I agree. I've always believed that the difference between
a mainframe and a mini had nothing to do with the size of
the box, but everything to do with architecture, particularly
I/O. Mainframes were record-oriented, while minis were
byte-oriented.
The PDP-8 was anything but byte-oriented, it was a 12-bit machine.
Plus, DECTape was block-oriented.
On 04/11/2026 19:51, Charlie Gibbs wrote:
A mediocre assembly language programmer would not let lack of
readability stand in the way. And once you get into self-modifying
code there are all sorts of horrible things you can do.
IBM macro assembler is about as machine code as COBOL
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-04, John Levine <johnl@taugh.com> wrote:
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>>>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
I agree. I've always believed that the difference between
a mainframe and a mini had nothing to do with the size of
the box, but everything to do with architecture, particularly
I/O. Mainframes were record-oriented, while minis were
byte-oriented.
The PDP-8 was anything but byte-oriented, it was a 12-bit machine.
Plus, DECTape was block-oriented.
On 04/11/2026 23:14, Carlos E. R. wrote:
On 2026-11-04 22:47, Lawrence DrCOOliveiro wrote:My mathematical friend needs to use assembler routines to access some of
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me >>>>>> they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written >>> in C, with very little assembly language.
Well, I know for a fact that this machine doesn't run any standard OS.
I asked him. Also, the disks do not have a filesystem, either (I also
asked). Of course most of the programmers work in some high level
language, but there are certain critical operations that are optimized
in assembler. I don't remember which, and obviously I'm not going to
quote him breaking confidence. But if anyone is a good programmer in
those machines, in assembler, they are hiring.
the vector processing that the chipset, but not available-a C libraries, supports.
IBM Macro-a assembler is practically a high level language, given the
scale of macro usage in legacy systems
On 2026-11-05, Thomas Koenig <tkoenig@netcologne.de> wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-04, John Levine <johnl@taugh.com> wrote:
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>>>>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
I agree. I've always believed that the difference between
a mainframe and a mini had nothing to do with the size of
the box, but everything to do with architecture, particularly
I/O. Mainframes were record-oriented, while minis were
byte-oriented.
The PDP-8 was anything but byte-oriented, it was a 12-bit machine.
Plus, DECTape was block-oriented.
You've got me there. It's amazing how slippery definitions become,
even when starting from "I know one when I see one." I could say
that minis tend to process a character at a time (e.g. a PDP-11
driving a VT50) while mainframes tend to process a record at a
time (e.g. a 360 filling a 2260's screen in a single operation).
But exceptions (e.g. your DECTape) creep in...
I do admit that the other extreme can become hairy. I once saw a
recursive macro that solved the Towers of Hanoi. It generated no
code; the solution was in the expanded source listing.
I usually just felt like mini=16 bits, mainframe=32 bits or more.
The 68000 had a beautiful assembler. Or rather, a beautiful
instruction set.
On Thu, 05 Nov 2026 21:57:51 GMT, Charlie Gibbs wrote:
I do admit that the other extreme can become hairy. I once saw a
recursive macro that solved the Towers of Hanoi. It generated no
code; the solution was in the expanded source listing.
String-based macro languages in general can be pretty hairy.
Towers of Hanoi wouldnrCOt even be that hard.
----
dnl
dnl Example m4 implementation of the GPM example output shown on page 64
dnl of Bryan Higman's "A Comparative Study of Programming Languages", 1st ed. dnl
dnl Created 2008 September 10 by Lawrence D'Oliveiro.
dnl
define(`repeat', `ifelse($2, 0, `', $2, 1, `$1', `$1`'repeat(`$1', eval($2 - 1))')')dnl
dnl repeat(str, n) outputs n consecutive repetitions of str
define(`n', `regexp(` x one two three four five six seven eight nine ten', `^\'repeat(` [^ ]+', $1)` \([^ ]+\)', `\1')')dnl
dnl n(i) returns the word representation of i. Need to use the "repeat" macro dnl because m4's regexps don't support \{n\} repeat coefficients. define(`cap', `regexp(`$1', `^\(.\)\(.*\)$', `translit(`\1', `a-z', `A-Z')\2')')dnl
dnl cap(s) returns s with the first letter capitalized.
dnl
define(`nrverses', 10)dnl
define(`nmen', `cap(n($1)) ifelse($1, 1, man, men)')dnl
dnl nmen(i) outputs the phrase "<i> man/men"
define(`tomow', ` went to mow')dnl
define(`mow1', `tomow a meadow')dnl
define(`firstline', `nmen($1)`'tomow,mow1')dnl
dnl firstline(i) outputs the first line of verse i.
define(`restlines', `ifelse($1, 1, `nmen(1) and his dog,mow1', `nmen($1) restlines(eval($1 - 1))')')dnl
dnl restlines(i) outputs the remaining lines of verse i
define(`verse', `firstline($1)
restlines($1)')dnl
dnl verse(i) outputs the complete verse i
define(`verses', `verse($1)`'ifelse($1, nrverses, `', `
verses(eval($1 + 1))')')dnl
dnl verses(i) outputs the verses from i up to the last one.
dnl here we go...
verses(1)
Lawrence =?iso-8859-13?q?D=FFOliveiro?= <ldo@nz.invalid> writes:
On Wed, 04 Nov 2026 06:21:03 GMT, Charlie Gibbs wrote:
Mainframes were record-oriented, while minis were byte-oriented.
Maybe better to say "minis were character-oriented". Remember DEC had
18-bit and 12-bit minicomputer architectures before IBM even invented
the term "byte".
ITYM "IBM even adopted the term 'byte' from the telecom industry." [...]
On Thu, 5 Nov 2026 22:24:10 +0100, Carlos E. R. wrote:
The 68000 had a beautiful assembler. Or rather, a beautiful
instruction set.
Definitely inspired by the PDP-11. Pity it was big-endian. Also, the
division into rCLaddressrCY versus rCLdatarCY registers did seem to be a step backwards.
HererCOs one trivia question: the decrement-and-branch instruction would
fall out of the loop when the count reached -1, not 0 as the
equivalent instruction in nearly every other computer architecture
did. Why?
On the other side of the coin, someone at Univac decided that
the key to productivity was not so much a quick response time
as a consistent one - which led to their system inserting
delays if a response was ready too soon. :-p
Whereas I read that the term Byte is attributed to Werner Buchholz
who'd have coined the term in 1956 while working for IBM.
The PDP-8 was anything but byte-oriented, it was a 12-bit machine.
Plus, DECTape was block-oriented.
You've got me there. It's amazing how slippery definitions become,
even when starting from "I know one when I see one." I could say
that minis tend to process a character at a time (e.g. a PDP-11
driving a VT50) while mainframes tend to process a record at a
time (e.g. a 360 filling a 2260's screen in a single operation).
But exceptions (e.g. your DECTape) creep in...
I usually just felt like mini=16 bits, mainframe=32 bits or more.
On 11/5/26 16:17, Lawrence DrCOOliveiro wrote:
On Thu, 5 Nov 2026 22:24:10 +0100, Carlos E. R. wrote:
The 68000 had a beautiful assembler. Or rather, a beautiful
instruction set.
Definitely inspired by the PDP-11. Pity it was big-endian. Also, the
division into rCLaddressrCY versus rCLdatarCY registers did seem to be a step
backwards.
HererCOs one trivia question: the decrement-and-branch instruction would
fall out of the loop when the count reached -1, not 0 as the
equivalent instruction in nearly every other computer architecture
did. Why?
Big-endian is the only true way.
On 11/5/26 16:17, Lawrence DrCOOliveiro wrote:
On Thu, 5 Nov 2026 22:24:10 +0100, Carlos E. R. wrote:
The 68000 had a beautiful assembler. Or rather, a beautiful
instruction set.
Definitely inspired by the PDP-11. Pity it was big-endian.
Big-endian is the only true way.
On Thu, 5 Nov 2026 19:37:43 -0700, Peter Flass wrote:
On 11/5/26 16:17, Lawrence DrCOOliveiro wrote:
On Thu, 5 Nov 2026 22:24:10 +0100, Carlos E. R. wrote:
The 68000 had a beautiful assembler. Or rather, a beautiful
instruction set.
Definitely inspired by the PDP-11. Pity it was big-endian.
Big-endian is the only true way.
The 68000 is a good example of the problems with it.
I remember reading about an experiment done in an IBM shop where
they shoveled in enough hardware and tweaked the protocols so that
terminal users got sub-second response time. Apparently it increased throughput even more than one would expect by just scaling the
response time - users could develop a rhythm and not have their
minds wander while waiting for a response.
On the other side of the coin, someone at Univac decided that the
key to productivity was not so much a quick response time as a
consistent one - which led to their system inserting delays if a
response was ready too soon. :-p
IBM Macro assembler is practically a high level language, given the
scale of macro usage in legacy systems
On 2026-11-04, Johann 'Myrkraverk' Oskarsson <johann@myrkraverk.invalid> wrote:
And if you think the fruit vendor is better, you can rest easy, because
at one time they refused to update my card number. Actually refused.
After reaching the /right/ support person, that person could not help
figure out why the website told me the card number had been updated, but
then the TV application always tried to use the old card number.
Ah, you've been around the mulberry bush with Netflix too, I presume.
We recently found we couldn't log in to Netflix because they didn't
like our card number - and quoted the last four digits and date of
a card that expired seven years ago. I wonder how they had managed
to bill us the previous month.
I called their customer support line, and the droid became downright belligerent, claiming that it wasn't their fault but our bank's.
I wound up hanging up on him, and we cancelled our Netflix account.
On 11/3/26 20:30, c186282 wrote:
-a NOTICED - my principal bank - a LOT of transactions
-a come up with a "pending" label. Might take DAYS. This
-a is not some 3rd-tier bank either.
-a Two or more larger credit/debit card charges, not
-a sure HOW they 'compete' in that system. If too
-a near each other one MIGHT get bumped. Problem !
-a In short, even 'online/instant' banking may not
-a be THAT instant.
Some of this is communication between banks, and some is your bank
milking the float for as long as is legally allowed.
I got in serious trouble for this,
because the process scheduler assumed that 500 ms was "a long time", and
a program that was going to wait for a long time should be swapped out
of memory.
Carlos E. R. wrote:
The 68000 had a beautiful assembler. Or rather, a beautiful
instruction set.
Definitely inspired by the PDP-11. Pity it was big-endian. Also, the
division into rCLaddressrCY versus rCLdatarCY registers did seem to be a step backwards.
HererCOs one trivia question: the decrement-and-branch instruction would
fall out of the loop when the count reached -1, not 0 as the
equivalent instruction in nearly every other computer architecture
did. Why?
On Thu, 5 Nov 2026 22:24:10 +0100, Carlos E. R. wrote:
The 68000 had a beautiful assembler. Or rather, a beautiful
instruction set.
Definitely inspired by the PDP-11. Pity it was big-endian. Also, the
division into rCLaddressrCY versus rCLdatarCY registers did seem to be a step backwards.
HererCOs one trivia question: the decrement-and-branch instruction would
fall out of the loop when the count reached -1, not 0 as the
equivalent instruction in nearly every other computer architecture
did. Why?
Lawrence DrCOOliveiro <ldo@nz.invalid> writes:
Carlos E. R. wrote:
The 68000 had a beautiful assembler. Or rather, a beautiful
instruction set.
Definitely inspired by the PDP-11. Pity it was big-endian. Also, the
division into rCLaddressrCY versus rCLdatarCY registers did seem to be a step
backwards.
HererCOs one trivia question: the decrement-and-branch instruction would
fall out of the loop when the count reached -1, not 0 as the
equivalent instruction in nearly every other computer architecture
did. Why?
At the surface level, because DBcc is a rather bizarrely shaped
instruction, with the condition code applying to the whole instruction rather; operating with inverse sense; and branching unless the decrement results in a value of -1 (i.e. 0xFFFF). So thererCOs _two_ conditional
steps (unless the condition is a constant, e.g. the common use of
DBF). Page 4-90 in the PRM[1] shows the definition.
So thererCOs several things to explain. I donrCOt have access to the
thinking of the designers, but I think the following are plausible explanations:
1) The whole-instruction conditionality allows for early exits from
loops (depending of course how they are structured). So you get two
checks for a single instruction.
2) Using 0xFFFF for the sentinel value comes, microarchitecturally, rCyfor
freerCO because it just means checking the carry out from the
decrement, which is 1 in precisely the situation where the decrement
reaches -1. Using 0 as the sentintel would mean testing the whole
result value.
3) The inverse sense of the conditional. The PRM explains this one in
terms of rCyUNTILrCO loops, which were perhaps more popular in the
programming languages of the time than they are today.
[1] https://www.nxp.com/docs/en/reference-manual/M68000PRM.pdf
On 11/5/26 14:57, Charlie Gibbs wrote:--
On 2026-11-05, Thomas Koenig <tkoenig@netcologne.de> wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-04, John Levine <johnl@taugh.com> wrote:
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while
programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
I agree. I've always believed that the difference between
a mainframe and a mini had nothing to do with the size of
the box, but everything to do with architecture, particularly
I/O. Mainframes were record-oriented, while minis were
byte-oriented.
The PDP-8 was anything but byte-oriented, it was a 12-bit machine.
Plus, DECTape was block-oriented.
You've got me there. It's amazing how slippery definitions become,
even when starting from "I know one when I see one." I could say
that minis tend to process a character at a time (e.g. a PDP-11
driving a VT50) while mainframes tend to process a record at a
time (e.g. a 360 filling a 2260's screen in a single operation).
But exceptions (e.g. your DECTape) creep in...
I usually just felt like mini=16 bits, mainframe=32 bits or more.
On 04/11/2026 14:54, Peter Flass wrote:
On 11/3/26 15:15, Pancho wrote:Not for transactuon processing
If you want heavy usage, I would guess bank risk calcs, weather calcs,
lots of science applications, running on compute grids, (or whatever
the modern word is), blow this mainframe high i/o away.
I think back in the day the quants preferred Sun.
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
And hererCOs another tale: this was probably from the very earliest days
of computers, but the story I remember reading involves users entering
a query, and how long it took to get back a response. Apparently, if
the response came back *too* quickly, the users felt that the computer somehow wasnrCOt treating their question with the importance it
deserved.
On 11/5/26 21:37, Peter Flass wrote:
On 11/5/26 16:17, Lawrence DrCOOliveiro wrote:
On Thu, 5 Nov 2026 22:24:10 +0100, Carlos E. R. wrote:
The 68000 had a beautiful assembler. Or rather, a beautiful
instruction set.
Definitely inspired by the PDP-11. Pity it was big-endian. Also, the
division into rCLaddressrCY versus rCLdatarCY registers did seem to be a step
backwards.
HererCOs one trivia question: the decrement-and-branch instruction would >>> fall out of the loop when the count reached -1, not 0 as the
equivalent instruction in nearly every other computer architecture
did. Why?
Big-endian is the only true way.
Ah, "religion" ...
Either way works fine.
PREF 'big' just for obviousness, but it's
not necessary.
On 2026-11-06 00:17, Lawrence DrCOOliveiro wrote:
HererCOs one trivia question: the decrement-and-branch instruction
would fall out of the loop when the count reached -1, not 0 as the
equivalent instruction in nearly every other computer architecture
did. Why?
Mmm. I don't remember. A flag went up?
On 11/4/2026 11:51, Charlie Gibbs wrote:
On the other side of the coin, someone at Univac decided that
the key to productivity was not so much a quick response time
as a consistent one - which led to their system inserting
delays if a response was ready too soon. :-p
Back in my University Computer Center days around 1972, I was hacking
a local email feature into the EXEC-8 editor, and decided that less
variation in response times would be a good thing, so I added a 1/2
second wait before each editor command prompt. I got in serious trouble
for this, because the process scheduler assumed that 500 ms was "a long time", and a program that was going to wait for a long time should be swapped out of memory.
On 2026-11-06, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
And hererCOs another tale: this was probably from the very earliest days
of computers, but the story I remember reading involves users entering
a query, and how long it took to get back a response. Apparently, if
the response came back *too* quickly, the users felt that the computer
somehow wasnrCOt treating their question with the importance it
deserved.
I wrote a program that searches a large collection of files for
various data. I noticed the same thing: it ran so fast that
people weren't taking it seriously. It didn't help that the
search parameter screen was too easy to use, either.
(This is all from memory, itrCOs been a long while since I wrote any 68000-family assembler.)
On 2026-11-06, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
And hererCOs another tale: this was probably from the very earliest days
of computers, but the story I remember reading involves users entering
a query, and how long it took to get back a response. Apparently, if
the response came back *too* quickly, the users felt that the computer
somehow wasnrCOt treating their question with the importance it
deserved.
I wrote a program that searches a large collection of files for
various data. I noticed the same thing: it ran so fast that
people weren't taking it seriously. It didn't help that the
search parameter screen was too easy to use, either.
I guess that in this modern age, where complexity is equated
with sophistication, any sign of simplicity or efficiency
is suspect.
On Fri, 6 Nov 2026 14:30:35 +0100, Carlos E. R. wrote:
On 2026-11-06 00:17, Lawrence DrCOOliveiro wrote:
HererCOs one trivia question: the decrement-and-branch instruction
would fall out of the loop when the count reached -1, not 0 as the
equivalent instruction in nearly every other computer architecture
did. Why?
Mmm. I don't remember. A flag went up?
It was quite a clever feature, which I think ended up widely
misunderstood. Otherwise more architectures would have copied it.
The reason is, so that if the loop count was zero, you could just
fall out of the loop automatically, without having to do an explicit
test for that case.
Example skeleton loop: (remember, the loop count was only 16 bits):
move.w -2loop-count-+, dN ; dN = some convenient d-register
bra loop_bottom
loop_top:
... body of loop ...
loop_bottom:
dbra dN, loop_top
... here on loop end ...
Notice this will loop exactly -2loop-count-+ times, even if -2loop-count-+
is zero.
If you want a 32-bit loop count, it only took a few more instructions:
move.l -2loop-count-+, dN
bra loop_bottom
loop_top1:
swap dN
loop_top2:
... body of loop ...
loop_bottom:
dbra dN, loop_top2
swap dN
dbra dN, loop_top1
... here on loop end ...
(This is all from memory, itrCOs been a long while since I wrote any 68000-family assembler.)
On 2026-11-06, Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
On 2026-11-06, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
And hererCOs another tale: this was probably from the very earliest days >>> of computers, but the story I remember reading involves users entering
a query, and how long it took to get back a response. Apparently, if
the response came back *too* quickly, the users felt that the computer
somehow wasnrCOt treating their question with the importance it
deserved.
I wrote a program that searches a large collection of files for
various data. I noticed the same thing: it ran so fast that
people weren't taking it seriously. It didn't help that the
search parameter screen was too easy to use, either.
That's much the same like people not taking a pair of headphones
seriously if they are light; so Beats famously added hidden metal
weights to them to create a more premium feel.
I've heard that car manufacturers put padding into the doors
to give a satisfying thump rather than a clank.
The Natural Philosopher <tnp@invalid.invalid> schrieb:
IBM Macro assembler is practically a high level language, given the
scale of macro usage in legacy systems
High-level languages abstract away the underlying machine, which
I presume the Macro assembler does not.
Plus, does it do register allocation?
(OK, then Wirth's PL/360 creeps in, which does not do so either,
but then I would argue that that is a glorified assembler - it also
does not do operator precenence).
On 2026-11-07 02:22, Charlie Gibbs wrote:
I've heard that car manufacturers put padding into the doors
to give a satisfying thump rather than a clank.
Unless the various sounds in cars are created artificially!
(Not joking. - Inspect the automotive business job offers,
specifically sound designers and sound engineers, and also
inspect the information on the Net about where artificial
sound is actually synthesized and applied to, including the
doors.)
In alt.folklore.computers John Levine <johnl@taugh.com> wrote:
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
Actually, 360/30 nicely shows that there is not much technical
difference. Hardware of 360/30 has 60 bit instructions
but otherwise several characteristics of a mini: 8 bit data
word and fast interrupts. ....
Also 360/44 may cause trouble. From I/O point of view it
is more mainframish than 360/30. But it lacks decimal
instructions and IIUC was intended to compete with bigger
minis from other manufacturers.
On 2026-11-07, Janis Papanagnou <janis_papanagnou+ng@hotmail.com> wrote:
On 2026-11-07 02:22, Charlie Gibbs wrote:
I've heard that car manufacturers put padding into the doors
to give a satisfying thump rather than a clank.
Unless the various sounds in cars are created artificially!
(Not joking. - Inspect the automotive business job offers,
specifically sound designers and sound engineers, and also
inspect the information on the Net about where artificial
sound is actually synthesized and applied to, including the
doors.)
I did read recently that they have staffers charged with
designing the weird noises electric cars make at low speeds.
The Natural Philosopher <tnp@invalid.invalid> writes:
On 04/11/2026 14:54, Peter Flass wrote:
On 11/3/26 15:15, Pancho wrote:Not for transactuon processing
If you want heavy usage, I would guess bank risk calcs, weather calcs, >>>> lots of science applications, running on compute grids, (or whatever
the modern word is), blow this mainframe high i/o away.
I think back in the day the quants preferred Sun.
You clearly don't know what a "quant" is, in this context.
I did read recently that they have staffers charged with
designing the weird noises electric cars make at low speeds.
ASM was also *the* way for the older micro-controllers
-a for a long time. However some of the optimized 'C'
-a compilers can now rival ASM for speed and size and
-a you don't have to hand-code bit-banging serial either
The 360/44 was intended for realtime process control. It had
a priority interrupt and a fast word at a time I/O feature. As
I said in another message, I don't think they sold very many of them.
The 1800, a sixteen bit mini that was a souped up 1130, was more
widely used for realtime work.
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do it
in 'C'.
On 11/7/26 04:45, The Natural Philosopher wrote:
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least). When PCs were new I used a compiler called "Aztec C" that had almost no overhead in excess libraries, start-
up code, etc., unless you asked for them.
On 11/7/2026 10:37 PM, Peter Flass wrote:
On 11/7/26 04:45, The Natural Philosopher wrote:
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least). When PCs were new I used a compiler
called "Aztec C" that had almost no overhead in excess libraries, start-
up code, etc., unless you asked for them.
And nonconsensual AVX code.
I've heard some CPUs have a lower maximum
Hz when AVX is enabled.
On 11/7/26 04:45, The Natural Philosopher wrote:
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least). When PCs were new I used a compiler called "Aztec C" that had almost no overhead in excess libraries, start-
up code, etc., unless you asked for them.
On 11/7/26 09:37, Peter Flass wrote:
On 11/7/26 04:45, The Natural Philosopher wrote:
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least). When PCs were new I used a compiler
called "Aztec C" that had almost no overhead in excess libraries,
start- up code, etc., unless you asked for them.
-a Used Aztec ... indeed HAVE the compiler and EXTENSIVE
-a docs stashed away somewhere. Will have to make a new
-a CP/M-86 VM for it first. Not sure you'd get best effect
-a on a simulated Z80. Have seen some 'kit' Z80 boards.
-a There's also a DOS version - but it's apparently meant
-a to run under XP.
-a The docs are IMPRESSIVE - endless bit-level and flag
-a stuff - exactly what you needed to know if trying to
-a shoehorn something into a small system.
-a A few versions of Turbo Pascal allowed code overlays
-a in a fairly civil fashion, another way to squeeze a
-a big pgm into a small machine. But, better have a HDD,
-a pulling overlays from a floppy - you don't wanna
-a go there :-) Did a couple of large pgms where overlays
-a were needed.
-a Anyway, if interested in Aztec, try :
https://www.aztecmuseum.ca/compilers.htm
I said in another message, I don't think they sold very many of them.
The 1800, a sixteen bit mini that was a souped up 1130, was more
widely used for realtime work.
Actually I think it was the other way - the 1130 was a stripped-down 1800.
But then you add all the cruft that GCC drags in ...
On 11/7/26 04:45, The Natural Philosopher wrote:
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do it
in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least). When PCs were new I used a compiler called "Aztec C" that had almost no overhead in excess libraries,
start-up code, etc., unless you asked for them.
On 11/7/26 04:45, The Natural Philosopher wrote:Hardly.
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least).
When PCs were new I used a compiler
called "Aztec C" that had almost no overhead in excess libraries,
start-up code, etc., unless you asked for them.
On 07/11/2026 14:37, Peter Flass wrote:
On 11/7/26 04:45, The Natural Philosopher wrote:Hardly.
On 07/11/2026 04:09, c186282 wrote:But then you add all the cruft that GCC drags in and you're back where
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C' compilers can
-a-a now rival ASM for speed and size and you don't have to hand-code
-a-a bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
you started (size-wise, at least).
When PCs were new I used a compiler called "Aztec C" that had almost noWell in general you DO ask for them
overhead in excess libraries, start-up code, etc., unless you asked for
them.
The 'mainframe' era is nearly over - to be replaced
by cluster/distributed/AI.
However IBM is already into all that, WILL make money.
Many baskets ... How To Survive/Thrive.
Hmmmm ... do remember when IBM was "typewriters" ...
1988, Nick Donofrio approves HA/6000, originally for NYTimes to move
their newspaper system ("ATEX") off DEC VAXCluster to RS/6000 (running project out at Los Gatos lab). I rename it HA/CMP https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing when I start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with RDBMS vendors (Oracle, Sybase, Ingres, Informix) with VAXCluster support in
same source base with UNIX (planning on using Hursley 9333 for mid-range
and full FCS for both technical and commercial scaleup). DB2 was still
all mainframe source, IBM non-mainframe "DB2" had yet to be developed.
IBM S/88 (relogo'ed Stratus) Product Administrator started taking us
around to their customers and also had me write a section for the
corporate continuous availability strategy document (it gets pulled when
both AS400/Rochester and mainframe/POK complain they couldn't meet requirements). Had coined "disaster survivability" and "geographic survivability" (as counter to disaster/recovery) when out marketing
HA/CMP.
Around 2001, my group had developed a T1-over-UDP-over-Ethernet bridge
to use in tandem with our wireless ethernet bridge product. After we
spun out from WiLan, and became Afar, we also marketed it
separately. Our largest sale was to NYT, to replace the dual T1 lines
they had to their (16?) regional printing plants, connected to the phototypesetters in the print shops. At 42nd street they had a rack
with 60(?) of our boxes. While they wanted the two lines for each
remote site to be on separate boxes connected to separate IP networks,
two lines to different sites could share one of our units, which had
two T1 jacks.
System worked well; I once got to visit the Old Lady shortly before
they moved down the street.
Latest public benchmark I've found is 2010 z196 "Peak I/O" getting 2M
IOPS using 104 FICON (20K IOPS/FICON). About the same time a native FCS
is announced for E5-2600 server blades getting over million IOPS (two
such FCS higher throughput than 104 FICON running over FCS). Note IBM
docs has SAPs (system assist processors that do actual I/O) CPU be kept
to 70% ... or 1.5M IOPS ... also no CKD DASD have been made for decades
(just simulated with industry standard fixed-block devices).
On 11/8/2026 2:31 AM, c186282 wrote:Seems it should; it was getting it to work under XP that Bill Buckels had troubles and has provided some extras, according to the readme.
On 11/7/26 09:37, Peter Flass wrote:
On 11/7/26 04:45, The Natural Philosopher wrote:
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
aa for a long time. However some of the optimized 'C'
aa compilers can now rival ASM for speed and size and
aa you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better. >>>
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do >>> it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least). When PCs were new I used a compiler
called "Aztec C" that had almost no overhead in excess libraries,
start- up code, etc., unless you asked for them.
a Used Aztec ... indeed HAVE the compiler and EXTENSIVE
a docs stashed away somewhere. Will have to make a new
a CP/M-86 VM for it first. Not sure you'd get best effect
a on a simulated Z80. Have seen some 'kit' Z80 boards.
a There's also a DOS version - but it's apparently meant
a to run under XP.
a The docs are IMPRESSIVE - endless bit-level and flag
a stuff - exactly what you needed to know if trying to
a shoehorn something into a small system.
a A few versions of Turbo Pascal allowed code overlays
a in a fairly civil fashion, another way to squeeze a
a big pgm into a small machine. But, better have a HDD,
a pulling overlays from a floppy - you don't wanna
a go there :-) Did a couple of large pgms where overlays
a were needed.
a Anyway, if interested in Aztec, try :
https://www.aztecmuseum.ca/compilers.htm
Thank you. It's mildly unclear if the compiler will run under pure DOS,
and this XP package is merely a convenience -- alas I was not in the
mood to find out.
... from which I learned that Stratus and Tandem both had figured
out how to do rolling upgrades in a non-stop environment; I had
thought that was an IBM mainframe invention.
Many techies despise System/Z as an overpriced dinosaur with an
obsolete architechture. But I wonder if it is any less expensive to
dip the Intel processors in that magic sauce. I bet IBM has better
scale to their builds.
There was then industry articles that server chip makers were
shipping half their products directly to megadatacenters. Shortly
later, IBM sells off its server blade business.
There was a paper on how Google did it, published 10 or 20 years ago.
At that point, they had a total complement of about 460,000 x86
servers. They would buy the cheapest component they could for every
component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are broken
at any moment? The users certainly wonrCOt notice.
Mainframes stopped being relevant back in the 1980s; every further development since then has just been driven by customer inertia.
Which kind of begs the question, doesnrCOt it: who bought it, and why?
If we define mainframes a large, very robust systems, they are still
very relevant. Although it may be true that the major technology was
in place before 1990 (non-stop cluster systems, transaction locking
with completion/rollback and strategic redundancy). What you call
customer inertia is also known as vendor lock-in.
On 11/8/2026 13:31, Lawrence DrCOOliveiro wrote:
There was a paper on how Google did it, published 10 or 20 years
ago. At that point, they had a total complement of about 460,000
x86 servers. They would buy the cheapest component they could for
every component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are
broken at any moment? The users certainly wonrCOt notice.
The parallelism of a Google is a very different beast from the
robust parallelism of a banking mainframe.
That is very different from transactions that must execute exactly
once.
Mainframes stopped being relevant back in the 1980s; every further
development since then has just been driven by customer inertia.
If we define mainframes a large, very robust systems, they are still
very relevant.
On Sun, 8 Nov 2026 05:59:55 -0800, Lars Poulsen wrote:
... from which I learned that Stratus and Tandem both had figured
out how to do rolling upgrades in a non-stop environment; I had
thought that was an IBM mainframe invention.
The IBM mega-marketing machine strikes again.
Or past tense, really; IBM no longer has that power. Microsoft
inherited that mantle, but it, too, is losing its grip.
Many techies despise System/Z as an overpriced dinosaur with an
obsolete architechture. But I wonder if it is any less expensive to
dip the Intel processors in that magic sauce. I bet IBM has better
scale to their builds.
There was a paper on how Google did it, published 10 or 20 years ago.
At that point, they had a total complement of about 460,000 x86
servers. They would buy the cheapest component they could for every
component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are broken
at any moment? The users certainly wonrCOt notice.
Mainframes stopped being relevant back in the 1980s; every further development since then has just been driven by customer inertia.
On Sun, 08 Nov 2026 08:32:30 -1000, Lynn Wheeler wrote:
There was then industry articles that server chip makers were
shipping half their products directly to megadatacenters. Shortly
later, IBM sells off its server blade business.
Which kind of begs the question, doesnrCOt it: who bought it, and why?
On Sun, 8 Nov 2026 13:45:03 -0800, Lars Poulsen wrote:
On 11/8/2026 13:31, Lawrence DrCOOliveiro wrote:
There was a paper on how Google did it, published 10 or 20 years
ago. At that point, they had a total complement of about 460,000
x86 servers. They would buy the cheapest component they could for
every component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are
broken at any moment? The users certainly wonrCOt notice.
The parallelism of a Google is a very different beast from the
robust parallelism of a banking mainframe.
Mainframes have never been that rCLrobustrCY. YourCOve mentioned elsewhere learning about Stratus and Tandem, who took robustness to a new level
back in the 1980s, beyond anything the mainframe world could match.
Think of that kind of redundancy multiplied tenfold, a hundredfold, a thousandfold, all built on open-source technologies: thatrCOs the kind
of world we live in today.
On 11/8/26 16:51, Lawrence DrCOOliveiro wrote:
On Sun, 8 Nov 2026 13:45:03 -0800, Lars Poulsen wrote:
On 11/8/2026 13:31, Lawrence DrCOOliveiro wrote:
There was a paper on how Google did it, published 10 or 20 years
ago. At that point, they had a total complement of about 460,000
x86 servers. They would buy the cheapest component they could for
every component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are
broken at any moment? The users certainly wonrCOt notice.
The parallelism of a Google is a very different beast from the
robust parallelism of a banking mainframe.
Mainframes have never been that rCLrobustrCY. YourCOve mentioned elsewhere >> learning about Stratus and Tandem, who took robustness to a new level
back in the 1980s, beyond anything the mainframe world could match.
Think of that kind of redundancy multiplied tenfold, a hundredfold, a
thousandfold, all built on open-source technologies: thatrCOs the kind
of world we live in today.
Mainframes typically achieve an uptime percentage between 99.999% (five nines) and 99.9999999% (nine nines)
Mainframes typically achieve an uptime percentage between 99.999%
(five nines) and 99.9999999% (nine nines)
On Sun, 8 Nov 2026 20:33:45 -0700, Peter Flass wrote:
Mainframes typically achieve an uptime percentage between 99.999%
(five nines) and 99.9999999% (nine nines)
The hyperscalers probably manage that well, if not better.
There is an article from 1986 on Bitsavers which describes having to
reboot an IBM mainframe to turn daylight saving on and off.
Mainframes typically achieve an uptime percentage between 99.999% (five nines) and 99.9999999% (nine nines)
On Sun, 8 Nov 2026 13:45:03 -0800, Lars Poulsen wrote:
On 11/8/2026 13:31, Lawrence DrCOOliveiro wrote:
There was a paper on how Google did it, published 10 or 20 years
ago. At that point, they had a total complement of about 460,000
x86 servers. They would buy the cheapest component they could for
every component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are
broken at any moment? The users certainly wonrCOt notice.
The parallelism of a Google is a very different beast from the
robust parallelism of a banking mainframe.
Mainframes have never been that rCLrobustrCY
learning about Stratus and Tandem, who took robustness to a new level
back in the 1980s, beyond anything the mainframe world could match.
Noted in this group though - I/O arch diffs between
mainframes and minis.
Noted in this group though - I/O arch diffs between
mainframes and minis.
Will someone try to elaborate ?
According to c186282 <c186282@nnada.net>:
Noted in this group though - I/O arch diffs between
mainframes and minis.
Will someone try to elaborate ?
The classic difference is that mainframes have channels, so the
processor sets up a bunch of work for the channel to do, the channel
goes off and does it independently of the CPU. Minis had much less
complex I/O controllers so they do more detailed stuff. An example I
gave a few messages back was for a card reader. A mainframe starts a
channel program, it reads in a card or maybe seveal cards, and reports
back with an interrupt. On a mini there was an I/O operation to start
the card reader, an interrupt for each column and an I/O to read it,
and then something at the end to stop the reader or go to the next
card.
But that was 40 or 50 years ago. These days even cheap generic disk controllers do more than channels did in the 1960s, e.g., keep a queue
of requests and schedule them to minimize delay, and handle spare block replacement for bad blocks.
Modern mainframes have extremely capable channels but the distinction is
more about reliability and things like hardware and software upgrades in place without rebooting.
Noted in this group though - I/O arch diffs between
mainframes and minis.
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
On 11/9/26 16:20, John Levine wrote:
According to c186282 <c186282@nnada.net>:
Noted in this group though - I/O arch diffs between
mainframes and minis.
Will someone try to elaborate ?
The classic difference is that mainframes have channels, so the
processor sets up a bunch of work for the channel to do, the channel
goes off and does it independently of the CPU. Minis had much less
complex I/O controllers so they do more detailed stuff. An example I
gave a few messages back was for a card reader. A mainframe starts a
channel program, it reads in a card or maybe seveal cards, and reports
back with an interrupt. On a mini there was an I/O operation to start
the card reader, an interrupt for each column and an I/O to read it,
and then something at the end to stop the reader or go to the next
card.
But that was 40 or 50 years ago. These days even cheap generic disk
controllers do more than channels did in the 1960s, e.g., keep a queue
of requests and schedule them to minimize delay, and handle spare block
replacement for bad blocks.
Modern mainframes have extremely capable channels but the distinction is
more about reliability and things like hardware and software upgrades in
place without rebooting.
Ok ... so 'mainframes' as you saw them were more
'directors' for distributed processing rather
than THE processor.
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to
take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY
setup that the big folks operate in your own homelab, because all the foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to
take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY--
setup that the big folks operate in your own homelab, because all the foundation software they use is available to you, too <https://docs.xcp-ng.org/vms/vm-migration/>.
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off- loaded to peripherials that were essentially limited computers of their
On 11/10/26 09:37, Alexander Schreiber wrote:
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off-
loaded to peripherials that were essentially limited computers of their
"Fully offloaded" is a bit weasel worded. I thought PCs did offload some duties to io devices, nic chip, wifi chip, gpu card. sound card. Also,
even a mainframe will need service routines to handle communications
with IO devices. Where to draw the line seems unclear.
Caveat: I don't have a clue about these things, it is just that what you
are saying doesn't sound quite right.
c186282 <c186282@nnada.net> wrote:
On 11/9/26 16:20, John Levine wrote:
According to c186282 <c186282@nnada.net>:
Noted in this group though - I/O arch diffs between
mainframes and minis.
Will someone try to elaborate ?
The classic difference is that mainframes have channels, so the
processor sets up a bunch of work for the channel to do, the channel
goes off and does it independently of the CPU. Minis had much less
complex I/O controllers so they do more detailed stuff. An example I
gave a few messages back was for a card reader. A mainframe starts a
channel program, it reads in a card or maybe seveal cards, and reports
back with an interrupt. On a mini there was an I/O operation to start
the card reader, an interrupt for each column and an I/O to read it,
and then something at the end to stop the reader or go to the next
card.
But that was 40 or 50 years ago. These days even cheap generic disk
controllers do more than channels did in the 1960s, e.g., keep a queue
of requests and schedule them to minimize delay, and handle spare block
replacement for bad blocks.
Modern mainframes have extremely capable channels but the distinction is >>> more about reliability and things like hardware and software upgrades in >>> place without rebooting.
Ok ... so 'mainframes' as you saw them were more
'directors' for distributed processing rather
than THE processor.
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off- loaded to peripherials that were essentially limited computers of their
own:
- CPU tells I/O interface (channel): "Do this I/O"
- CPU goes back to computing, channel does some I/O heavy lifting
- channel has completed I/O and sends an "Hey, I'm done" interrupt
to the CPU
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher
reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of
mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to
take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY
setup that the big folks operate in your own homelab, because all the
foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
Highly distributed systems like those run and offered by the hyperscalers
are built on an entirely different design philosophy:
- _everything_ fails all the time (law of big numbers: take the MTBF
for a compute component, run a lot of them and you've got a steady
rain of failures, e.g. say you've got an MTBF of 1 million hours and
you run 2 million of these devices, then on average you'll have a
failure every half hour)
On 11/10/26 09:37, Alexander Schreiber wrote:
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off-
loaded to peripherials that were essentially limited computers of their
"Fully offloaded" is a bit weasel worded. I thought PCs did offload some >duties to io devices, nic chip, wifi chip, gpu card. sound card. Also,
even a mainframe will need service routines to handle communications
with IO devices. Where to draw the line seems unclear.
Caveat: I don't have a clue about these things, it is just that what you--- Synchronet 3.22a-Linux NewsLink 1.2
are saying doesn't sound quite right.
c186282 <c186282@nnada.net> schrieb:
Noted in this group though - I/O arch diffs between
mainframes and minis.
I have a somewhat personal definition of a mainframe, which I also
includes the classical ones.
A mainframe is a computer that
- is very expensive
- takes large resources to run (a computer center)
- has very good performance in at least one field
- is not a cluster, massively parallel etc
On 11/10/26 02:50, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher
reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of
mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to >>> take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY
setup that the big folks operate in your own homelab, because all the
foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
Highly distributed systems like those run and offered by the hyperscalers
are built on an entirely different design philosophy:
- _everything_ fails all the time (law of big numbers: take the MTBF
for a compute component, run a lot of them and you've got a steady
rain of failures, e.g. say you've got an MTBF of 1 million hours and
you run 2 million of these devices, then on average you'll have a
failure every half hour)
Didn't I read somewhere that techs in large datacenters get around on >rollerblades to swap out failing components?
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off- loaded to peripherials that were essentially limited computers of their
own:
- CPU tells I/O interface (channel): "Do this I/O"
- CPU goes back to computing, channel does some I/O heavy lifting
- channel has completed I/O and sends an "Hey, I'm done" interrupt
to the CPU
Note: 1980, IBM STL (since renamed SVL) was bursting at seams and 300 people/terminals from IMS group were being moved to offsite bldg, with dataprocessing service back to STL datacenter. They had tried "remote
3270" but found human factors totally unacceptable.
I get con'ed into doing channel extender support, allowing placing 3270 controllers at offsite bldg with no perceived difference in human
factors. Unanticipated side-effect was the associated systems'
throughput increase 10-15 percent. STL had been speading 3270
controllers across all channels with 3330 drives. The channel-extenders (operating full-duplex) significantly reduced the channel busy
interference with the 3330 drives (for same amount of 3270 traffic;
there was some consideration moving all 3270 controllers to channel-extenders).
c186282 <c186282@nnada.net> wrote:
Ok ... so 'mainframes' as you saw them were more
'directors' for distributed processing rather
than THE processor.
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off- loaded to peripherials that were essentially limited computers of their
own:
- CPU tells I/O interface (channel): "Do this I/O"
- CPU goes back to computing, channel does some I/O heavy lifting
- channel has completed I/O and sends an "Hey, I'm done" interrupt
to the CPU
A prime example of neat code is that HASP (now JES) printed an entire
page with one channel program instead of a line at a time. It also read
a number of cards with one channel program instead of one card at a
time, or, as someone has mentioned, one card column at a time.
I read that, when minis came along, their first defining
characteristic was that they didn't need computer center resources
and fit in a single room.
Alexander Schreiber <als@usenet.thangorodrim.de> writes:
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off-
loaded to peripherials that were essentially limited computers of their
own:
- CPU tells I/O interface (channel): "Do this I/O"
- CPU goes back to computing, channel does some I/O heavy lifting
- channel has completed I/O and sends an "Hey, I'm done" interrupt
to the CPU
Original 360, there was limited processor and storage, but a lof of
excess channel capacity ... as a result things like finding data was offloaded to channels ... which would "search" for desired data ... and
as a result, channels ran half-duplex ... with channel dedicated to
single operation ... constantly pulling search argument from processor storage.
Move into 80s, storage got lot more abundant ... and channel
architecture increasingly channel I/O increasingly became
system throughput bottleneck.
Early 80s, I wrote tome that 60s era 360/67 could get 70-80 CP67/CMS
users on system ... and early 80s, with IBM processors getting 40-50
times faster and storage increasing 40-50 larger, should be able be see 3000-4000 users, but was only seeing 300-400 users (because disks &
channels throughput only increased 3-5 times (partly because half-duplex operation increasingly played a latency role in fully being able to see benefits of any speed increase).
Note once the 3033 was out the door, the processor engineers start on trout/3090. They configure memory, processor, channels, I/O to meet
specific throughput goals. However, in San Jose, the administration
dictated that the 3880 disk controller would use a (really slow,
resulting in significant channel busy) vertical CISC processor (much
slower than the previous processor in 3830 controller).
The 3090 people thought the 3880 would be the same as 3830, but
supporting 3380 3mbyte/sec transfer, when they found out how bad the
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but
they have extremely high reliability (you can literally fire a
bullet through a CPU, and things will migrate smoothly - if you
don't hit the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much
higher reliability through massive parallelism, based on cheap, yet
highly redundant hardware running a Free Software stack that is not
priced per CPU. That is much more cost-effective and any equivalent
number of mainframes. And there is no rCLRecovery UnitrCY that you can
shoot at to take it out.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
On 2026-11-10, Thomas Koenig <tkoenig@netcologne.de> wrote:
I read that, when minis came along, their first defining
characteristic was that they didn't need computer center resources
and fit in a single room.
And you could plug them into a wall socket.
"Fully offloaded" is a bit weasel worded. I thought PCs did offload some
duties to io devices, nic chip, wifi chip, gpu card. sound card. Also,
even a mainframe will need service routines to handle communications
with IO devices. Where to draw the line seems unclear.
Fully offloaded is accurate. For the Burroughs mainframes, a single instruction -initiated- an I/O request to a channel. The IIO (later
SPIO) instruction would instruct the I/O processor (in the late 70s
the IOP used an 8085 uProcessor) to initiate an I/O by passing a
6-digit (24-bit) opcode/variant word and the start and end addresses
of the buffer. The IOP would pass the command to a Data-Link Processor (DLP) that handled communications to the device controller (magnetic
tape, disk) or unit record device (printer, card reader, card punch)
or external communications processor (RS232/20ma current loop) and wait
for the controller to start passing data back to the IOP. The IOP
would move the data to the appropriate memory buffer and, when complete, signal the processor that the I/O had completed with an interrupt.
Caveat: I don't have a clue about these things, it is just that what you
are saying doesn't sound quite right.
On 2026-11-09, Lawrence DrCOOliveiro wrote:
On Sun, 8 Nov 2026 20:33:45 -0700, Peter Flass wrote:
Mainframes typically achieve an uptime percentage between 99.999%
(five nines) and 99.9999999% (nine nines)
The hyperscalers probably manage that well, if not better.
There is an article from 1986 on Bitsavers which describes having to
reboot an IBM mainframe to turn daylight saving on and off.
*That* sounds like complicating a problem that would be made simpler by
not using the local time, or at least DST-adjusted local time.
On 11/10/26 09:37, Alexander Schreiber wrote:
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off-
loaded to peripherials that were essentially limited computers of their
"Fully offloaded" is a bit weasel worded. I thought PCs did offload some duties to io devices, nic chip, wifi chip, gpu card. sound card. Also,
even a mainframe will need service routines to handle communications
with IO devices. Where to draw the line seems unclear.
Caveat: I don't have a clue about these things, it is just that what you
are saying doesn't sound quite right.
On 11/10/26 02:50, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher
reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of
mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to >>> take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY
setup that the big folks operate in your own homelab, because all the
foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
Highly distributed systems like those run and offered by the hyperscalers
are built on an entirely different design philosophy:
- _everything_ fails all the time (law of big numbers: take the MTBF
for a compute component, run a lot of them and you've got a steady
rain of failures, e.g. say you've got an MTBF of 1 million hours and
you run 2 million of these devices, then on average you'll have a
failure every half hour)
Didn't I read somewhere that techs in large datacenters get around on rollerblades to swap out failing components?
I/O was not done by the CPU (like it was done on homeI see you never used the serial cards with up to 32 terminals attached
computers were all I/O was entirely handled by the CPU), but fully off- loaded to peripherials that were essentially limited computers of their
own:
On 11/10/26 09:37, Alexander Schreiber wrote:
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off-
loaded to peripherials that were essentially limited computers of their
"Fully offloaded" is a bit weasel worded.
I thought PCs did offload some duties to io devices, nic chip, wifi chip, gpu card. sound card.
Some minis offered an option for peripherals to have at least
some of this kind of smarts. It was called "DMA" (direct memory
access), ...
Didn't I read somewhere that techs in large datacenters get around on >rollerblades to swap out failing components?
... a key design aspect of mainframes back then was that I/O was not
done by the CPU (like it was done on home computers were all I/O was
entirely handled by the CPU), but fully off- loaded to peripherials
that were essentially limited computers of their own ...
Which leaves most of the CPU free to do compute work instead of
busily doing lots of I/O. This and other design concepts (like form
based terminals doing local editing - think web forms filled in the
browser, send as a whole form and processed as a single request by
the server, instead of the server handling every keystroke on the
terminal) enabled having thousands of interactive users on one
mainframe ...
A prime example of neat code is that HASP (now JES) printed an
entire page with one channel program instead of a line at a time.
On 11/10/26 10:55, Lynn Wheeler wrote:
Note once the 3033 was out the door, the processor engineers start on
trout/3090. They configure memory, processor, channels, I/O to meet
specific throughput goals. However, in San Jose, the administration
dictated that the 3880 disk controller would use a (really slow,
resulting in significant channel busy) vertical CISC processor (much
slower than the previous processor in 3830 controller).
The 3090 people thought the 3880 would be the same as 3830, but
supporting 3380 3mbyte/sec transfer, when they found out how bad the
Anyone would think that newer=faster. Who would have thought it would be slower?
Yes, you can give me a lot of details, but I'm not seeing the
fundamental difference between that and DMA IO devices that have been
around since the first IBM PC in 1981.
On 11/10/26 09:37, Alexander Schreiber wrote:
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off-
loaded to peripherials that were essentially limited computers of their
"Fully offloaded" is a bit weasel worded. I thought PCs did offload some duties to io devices, nic chip, wifi chip, gpu card. sound card. Also,
even a mainframe will need service routines to handle communications
with IO devices. Where to draw the line seems unclear.
Caveat: I don't have a clue about these things, it is just that what you
are saying doesn't sound quite right.
c186282 <c186282@nnada.net> wrote:
On 11/9/26 16:20, John Levine wrote:
According to c186282 <c186282@nnada.net>:
Noted in this group though - I/O arch diffs between
mainframes and minis.
Will someone try to elaborate ?
The classic difference is that mainframes have channels, so the
processor sets up a bunch of work for the channel to do, the channel
goes off and does it independently of the CPU. Minis had much less
complex I/O controllers so they do more detailed stuff. An example I
gave a few messages back was for a card reader. A mainframe starts a
channel program, it reads in a card or maybe seveal cards, and reports
back with an interrupt. On a mini there was an I/O operation to start
the card reader, an interrupt for each column and an I/O to read it,
and then something at the end to stop the reader or go to the next
card.
But that was 40 or 50 years ago. These days even cheap generic disk
controllers do more than channels did in the 1960s, e.g., keep a queue
of requests and schedule them to minimize delay, and handle spare block
replacement for bad blocks.
Modern mainframes have extremely capable channels but the distinction is >>> more about reliability and things like hardware and software upgrades in >>> place without rebooting.
Ok ... so 'mainframes' as you saw them were more
'directors' for distributed processing rather
than THE processor.
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off- loaded to peripherials that were essentially limited computers of their
own:
- CPU tells I/O interface (channel): "Do this I/O"
- CPU goes back to computing, channel does some I/O heavy lifting
- channel has completed I/O and sends an "Hey, I'm done" interrupt
to the CPU
Which leaves most of the CPU free to do compute work instead of busily
doing lots of I/O. This and other design concepts (like form based
terminals doing local editing - think web forms filled in the browser, send as a whole form and processed as a single request by the server, instead
of the server handling every keystroke on the terminal) enabled having thousands of interactive users on one mainframe, since the mainframe CPU
did almost entirely compute (e.g. run the database and software for inventory, accounting, ...) and all I/O was offloaded to specialized computers (channels) acting as I/O interfaces. You essentially pay with
extra complexity for a far more capable system.
Kind regards,
Alex.
Anyone would think that newer=faster. Who would have thought it
would be slower?
On 11/10/26 02:37, Alexander Schreiber wrote:
c186282 <c186282@nnada.net> wrote:
On 11/9/26 16:20, John Levine wrote:
According to c186282-a <c186282@nnada.net>:
-a-a-a Noted in this group though - I/O arch diffs between
-a-a-a mainframes and minis.
-a-a-a Will someone try to elaborate ?
The classic difference is that mainframes have channels, so the
processor sets up a bunch of work for the channel to do, the channel
goes off and does it independently of the CPU. Minis had much less
complex I/O controllers so they do more detailed stuff. An example I
gave a few messages back was for a card reader. A mainframe starts a
channel program, it reads in a card or maybe seveal cards, and reports >>>> back with an interrupt. On a mini there was an I/O operation to start
the card reader, an interrupt for each column and an I/O to read it,
and then something at the end to stop the reader or go to the next
card.
But that was 40 or 50 years ago.-a These days even cheap generic disk
controllers do more than channels did in the 1960s, e.g., keep a queue >>>> of requests and schedule them to minimize delay, and handle spare block >>>> replacement for bad blocks.
Modern mainframes have extremely capable channels but the
distinction is
more about reliability and things like hardware and software
upgrades in
place without rebooting.
-a-a-a Ok ... so 'mainframes' as you saw them were more
-a-a-a 'directors' for distributed processing rather
-a-a-a than THE processor.
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off-
loaded to peripherials that were essentially limited computers of their
own:
-a - CPU tells I/O interface (channel): "Do this I/O"
-a - CPU goes back to computing, channel does some I/O heavy lifting
-a - channel has completed I/O and sends an "Hey, I'm done" interrupt
-a-a-a to the CPU
A prime example of neat code is that HASP (now JES) printed an entire
page with one channel program instead of a line at a time. It also read
a number of cards with one channel program instead of one card at a
time, or, as someone has mentioned, one card column at a time.
On 11/10/26 02:50, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher
reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of
mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to >>> take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY
setup that the big folks operate in your own homelab, because all the
foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
Highly distributed systems like those run and offered by the hyperscalers
are built on an entirely different design philosophy:
-a - _everything_ fails all the time (law of big numbers: take the MTBF
-a-a-a for a compute component, run a lot of them and you've got a steady
-a-a-a rain of failures, e.g. say you've got an MTBF of 1 million hours and >> -a-a-a you run 2 million of these devices, then on average you'll have a
-a-a-a failure every half hour)
Didn't I read somewhere that techs in large datacenters get around on rollerblades to swap out failing components?
Pancho <Pancho.Jones@protonmail.com> writes:
On 11/10/26 09:37, Alexander Schreiber wrote:
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off-
loaded to peripherials that were essentially limited computers of their
"Fully offloaded" is a bit weasel worded. I thought PCs did offload some
duties to io devices, nic chip, wifi chip, gpu card. sound card. Also,
even a mainframe will need service routines to handle communications
with IO devices. Where to draw the line seems unclear.
Fully offloaded is accurate. For the Burroughs mainframes, a single instruction -initiated- an I/O request to a channel. The IIO (later
SPIO) instruction would instruct the I/O processor (in the late 70s
the IOP used an 8085 uProcessor) to initiate an I/O by passing a
6-digit (24-bit) opcode/variant word and the start and end addresses
of the buffer. The IOP would pass the command to a Data-Link Processor (DLP) that handled communications to the device controller (magnetic
tape, disk) or unit record device (printer, card reader, card punch)
or external communications processor (RS232/20ma current loop) and wait
for the controller to start passing data back to the IOP. The IOP
would move the data to the appropriate memory buffer and, when complete, signal the processor that the I/O had completed with an interrupt.
Caveat: I don't have a clue about these things, it is just that what you
are saying doesn't sound quite right.
On Tue, 10 Nov 2026 12:19:31 -0700, Peter Flass wrote:
Anyone would think that newer=faster. Who would have thought it
would be slower?
As another example, I recall it said that passenger airliners after
the Boeing 747 actually have slightly lower cruising speeds. I guess
itrCOs all part of the goal of saving fuel.
On 10/11/2026 09:37, Alexander Schreiber wrote:
I/O was not done by the CPU (like it was done on homeI see you never used the serial cards with up to 32 terminals attached
computers were all I/O was entirely handled by the CPU), but fully off-
loaded to peripherials that were essentially limited computers of their
own:
to a PC running SCO unix.
They were essentially IO processors
On 11/10/26 15:14, Scott Lurndal wrote:
"Fully offloaded" is a bit weasel worded. I thought PCs did offload some >>> duties to io devices, nic chip, wifi chip, gpu card. sound card. Also,
even a mainframe will need service routines to handle communications
with IO devices. Where to draw the line seems unclear.
Fully offloaded is accurate. For the Burroughs mainframes, a single
instruction -initiated- an I/O request to a channel. The IIO (later
SPIO) instruction would instruct the I/O processor (in the late 70s
the IOP used an 8085 uProcessor) to initiate an I/O by passing a
6-digit (24-bit) opcode/variant word and the start and end addresses
of the buffer. The IOP would pass the command to a Data-Link Processor
(DLP) that handled communications to the device controller (magnetic
tape, disk) or unit record device (printer, card reader, card punch)
or external communications processor (RS232/20ma current loop) and wait
for the controller to start passing data back to the IOP. The IOP
would move the data to the appropriate memory buffer and, when complete,
signal the processor that the I/O had completed with an interrupt.
Yes, you can give me a lot of details, but I'm not seeing the
fundamental difference between that and DMA IO devices that have been
around since the first IBM PC in 1981.
As I said I'm naive of this kind of low level stuff, but from my view as
an application programmer I'm failing to see a fundamental difference
from a PC, offloading an IO task to a non CPU device with communication
via registers, interrupts and DMA.
On Tue, 10 Nov 2026 10:37:49 +0100, Alexander Schreiber wrote:
... a key design aspect of mainframes back then was that I/O was not
done by the CPU (like it was done on home computers were all I/O was
entirely handled by the CPU), but fully off- loaded to peripherials
that were essentially limited computers of their own ...
Which made sense back when the CPU was a scarce and expensive resource.
It no longer made sense when that was no longer true.
Which leaves most of the CPU free to do compute work instead of
busily doing lots of I/O. This and other design concepts (like form
based terminals doing local editing - think web forms filled in the
browser, send as a whole form and processed as a single request by
the server, instead of the server handling every keystroke on the
terminal) enabled having thousands of interactive users on one
mainframe ...
Nowadays we have millions of users on social media simultaneously, but
that request/response paradigm is no longer sufficient for the
real-time interaction requirements.
On 11/10/26 09:37, Peter Flass wrote:
On 11/10/26 02:50, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher
reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of >>>> mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to >>>> take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY >>>> setup that the big folks operate in your own homelab, because all the
foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
Highly distributed systems like those run and offered by the hyperscalers >>> are built on an entirely different design philosophy:
-a - _everything_ fails all the time (law of big numbers: take the MTBF
-a-a-a for a compute component, run a lot of them and you've got a steady >>> -a-a-a rain of failures, e.g. say you've got an MTBF of 1 million hours and >>> -a-a-a you run 2 million of these devices, then on average you'll have a >>> -a-a-a failure every half hour)
Didn't I read somewhere that techs in large datacenters get around on
rollerblades to swap out failing components?
IS an issue !!!
Linux/UNIX are 'better' at this.
As for 'failures every half hour' ... can't
HAVE that !
According to Peter Flass <Peter@Iron-Spring.com>:
Didn't I read somewhere that techs in large datacenters get around on >>rollerblades to swap out failing components?
I don't think so. I've read that Google has people wandering around with grocery carts full of spare parts that they swap into failed servers.
On 2026-11-11, Lawrence DrCOOliveiro wrote:
On Tue, 10 Nov 2026 12:19:31 -0700, Peter Flass wrote:
Anyone would think that newer=faster. Who would have thought it
would be slower?
As another example, I recall it said that passenger airliners
after the Boeing 747 actually have slightly lower cruising speeds.
I guess itrCOs all part of the goal of saving fuel.
It is possible if the 747 was inefficient. IIRC there's a sweet spot
(or range?) for efficient fuel usage in subsonic jets.
(For supersonic, IIRC Aerospatiale-BAC Concorde was actually quite
efficient in cruise (compared to other supersonic aircraft), and I
think the proposed redesign would even do away with the need for wet
thrust (necessary for TO and transonic climb?).)
The Natural Philosopher <tnp@invalid.invalid> wrote:
On 10/11/2026 09:37, Alexander Schreiber wrote:
I/O was not done by the CPU (like it was done on homeI see you never used the serial cards with up to 32 terminals attached
computers were all I/O was entirely handled by the CPU), but fully off-
loaded to peripherials that were essentially limited computers of their
own:
to a PC running SCO unix.
I never touched SCO Unix, I got my Unix start on SunOS/Solaris and very
early Linux.
They were essentially IO processors
Sure, they have to be to handle than many serial lines.
But even in the "PC server" space, those were not that common.
Kind regards,--
Alex.
Does not compare. In the PC case, the controller with DMA handles simple requests such as "fetch X blocks of data from address Y on block deviceI never did channel programming at the SIO instruction level, but a
Z". For mainframe channel systems, while they also do this, they can do
a lot more since they can run dedicated programs send to them as part of
the request - including processing the data beyond "get me some bytes".
Other folks already mentioned examples like telling the channel controller
to go grab a specific file from the filesystem on an attached block device.
On a PC, only the actual fetching of disk blocks would be offloaded to
the controller and the filesystem handling (reading the start of the file- system, traversing the file system structures to find the list of blocks occupied by the file and its size) would have been done by the CPU.
Pancho <Pancho.Jones@protonmail.com> wrote:
On 11/10/26 15:14, Scott Lurndal wrote:
"Fully offloaded" is a bit weasel worded. I thought PCs did offload some >>>> duties to io devices, nic chip, wifi chip, gpu card. sound card. Also, >>>> even a mainframe will need service routines to handle communications
with IO devices. Where to draw the line seems unclear.
Fully offloaded is accurate. For the Burroughs mainframes, a single
instruction -initiated- an I/O request to a channel. The IIO (later
SPIO) instruction would instruct the I/O processor (in the late 70s
the IOP used an 8085 uProcessor) to initiate an I/O by passing a
6-digit (24-bit) opcode/variant word and the start and end addresses
of the buffer. The IOP would pass the command to a Data-Link Processor >>> (DLP) that handled communications to the device controller (magnetic
tape, disk) or unit record device (printer, card reader, card punch)
or external communications processor (RS232/20ma current loop) and wait
for the controller to start passing data back to the IOP. The IOP
would move the data to the appropriate memory buffer and, when complete, >>> signal the processor that the I/O had completed with an interrupt.
Yes, you can give me a lot of details, but I'm not seeing the
fundamental difference between that and DMA IO devices that have been
around since the first IBM PC in 1981.
First of all, those capabilities where standard - to the point where DMA wasn't mentioned as a feature because it was an assumed part of the tech baseline - in mainframes long before the 1980s.
As I said I'm naive of this kind of low level stuff, but from my view as
an application programmer I'm failing to see a fundamental difference
from a PC, offloading an IO task to a non CPU device with communication
via registers, interrupts and DMA.
Does not compare. In the PC case, the controller with DMA handles simple requests such as "fetch X blocks of data from address Y on block device
Z". For mainframe channel systems, while they also do this, they can do
a lot more since they can run dedicated programs send to them as part of
the request - including processing the data beyond "get me some bytes".
Other folks already mentioned examples like telling the channel controller
to go grab a specific file from the filesystem on an attached block device.
On a PC, only the actual fetching of disk blocks would be offloaded to
the controller and the filesystem handling (reading the start of the file- system, traversing the file system structures to find the list of blocks occupied by the file and its size) would have been done by the CPU.
Kind regards,
Alex.
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 10:37:49 +0100, Alexander Schreiber wrote:
... a key design aspect of mainframes back then was that I/O was not
done by the CPU (like it was done on home computers were all I/O was
entirely handled by the CPU), but fully off- loaded to peripherials
that were essentially limited computers of their own ...
Which made sense back when the CPU was a scarce and expensive resource.
It no longer made sense when that was no longer true.
It still makes sense and that is borne out in hardware design. We haven't flipped back to polled I/O because "CPUs are faster than we know what to
do with", quite the opposite: modern GPUs offload most the fancy 3D graphics processing (originally, these days it's also a lot of matrix math for AI), modern NICs offer a lot of offload (modern DC NICs are full on computers running Linux and e.g. handle network traffic encryption on their own,
see e.g. NVidia Bluefield DPUs) and so on.
Which leaves most of the CPU free to do compute work instead of
busily doing lots of I/O. This and other design concepts (like form
based terminals doing local editing - think web forms filled in the
browser, send as a whole form and processed as a single request by
the server, instead of the server handling every keystroke on the
terminal) enabled having thousands of interactive users on one
mainframe ...
Nowadays we have millions of users on social media simultaneously, but
that request/response paradigm is no longer sufficient for the
real-time interaction requirements.
The "Unix serial terminal" model never applied to web interactive traffic, that was always running on the "smart terminal doing local edits - mainframe style" model. Mostly because that's a lot easier to implement and also it scales a lot better.
Kind regards,
Alex.
I never did channel programming at the SIO instruction level, but a colleague who did, told me that once mounted, you could copy one tape to another with a single SIO instruction, including backspace-retry on read
and write errors.
My direct experience with channel programming relates to programming a
Z80 on a UNIBUS card to simulate reads and writes on a selector channel
in order for a VAX to drive a line printer (really a Xerox 9700 laser printer - Paul Kyzivat of DEC write the VMS symbiont for it).
On 11/10/26 04:37, Alexander Schreiber wrote:<snip>
c186282 <c186282@nnada.net> wrote:
On 11/9/26 16:20, John Levine wrote:
According to c186282 <c186282@nnada.net>:
Noted in this group though - I/O arch diffs between
mainframes and minis.
Will someone try to elaborate ?
Which leaves most of the CPU free to do compute work instead of busily
doing lots of I/O. This and other design concepts (like form based
terminals doing local editing - think web forms filled in the browser, send >> as a whole form and processed as a single request by the server, instead
of the server handling every keystroke on the terminal) enabled having
thousands of interactive users on one mainframe, since the mainframe CPU
did almost entirely compute (e.g. run the database and software for
inventory, accounting, ...) and all I/O was offloaded to specialized
computers (channels) acting as I/O interfaces. You essentially pay with
extra complexity for a far more capable system.
Kind regards,
Alex.
I kind of see what model you're describing.
Really not THAT much different from what
I outlined.
"Smart" peripherials are good. CPU doesn't have
to do nearly as much. The downside is that now
you're hostage to the 'IQ' in the peripherials
rather than fine managing/optimizing via the
main CPU.
On 11/10/26 09:35, Peter Flass wrote:
On 11/10/26 02:37, Alexander Schreiber wrote:
A prime example of neat code is that HASP (now JES) printed an entire
page with one channel program instead of a line at a time. It also read
a number of cards with one channel program instead of one card at a
time, or, as someone has mentioned, one card column at a time.
"Smart" periphs are good ... but also
a kind of trap.
Ideally the CPU does almost everything - so you
can optimize/tweak.
But esp in the Olde Days the CPU really wasn't
that strong or had that many resources.
Pancho <Pancho.Jones@protonmail.com> wrote:
On 11/10/26 15:14, Scott Lurndal wrote:
Does not compare. In the PC case, the controller with DMA handles simple >requests such as "fetch X blocks of data from address Y on block device
Z". For mainframe channel systems, while they also do this, they can do
a lot more since they can run dedicated programs send to them as part of
the request - including processing the data beyond "get me some bytes".
On 11/11/2026 11:03, Alexander Schreiber wrote:
Depends on your age and experience. I saw many. A multiuser system that >could handle 32 screens running a database oriented program for <
But even in the "PC server" space, those were not that common.
$10,000 was a very good alternative to a VAX.
Adding in TCP/IP rendered the serial ports obsolete and made it even
better, apart from the cost per desktop of a PC running
telnet...compared to a VT220 or a Wyse 50.
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 10:37:49 +0100, Alexander Schreiber wrote:
... a key design aspect of mainframes back then was that I/O was not
done by the CPU (like it was done on home computers were all I/O was
entirely handled by the CPU), but fully off- loaded to peripherials
that were essentially limited computers of their own ...
Which made sense back when the CPU was a scarce and expensive resource.
It no longer made sense when that was no longer true.
It still makes sense and that is borne out in hardware design. We haven't >flipped back to polled I/O because "CPUs are faster than we know what to
do with", quite the opposite: modern GPUs offload most the fancy 3D graphics >processing (originally, these days it's also a lot of matrix math for AI), >modern NICs offer a lot of offload (modern DC NICs are full on computers >running Linux and e.g. handle network traffic encryption on their own,
see e.g. NVidia Bluefield DPUs) and so on.
I see you never used the serial cards with up to 32 terminals
attached to a PC running SCO unix.
I never touched SCO Unix, I got my Unix start on SunOS/Solaris and
very early Linux.
They were essentially IO processors
Sure, they have to be to handle than many serial lines.
But even in the "PC server" space, those were not that common.
Depends on your age and experience. I saw many. A multiuser system
that could handle 32 screens running a database oriented program for
< $10,000 was a very good alternative to a VAX.
On Wed, 11 Nov 2026 12:36:43 +0000
The Natural Philosopher <tnp@invalid.invalid> wrote:
I see you never used the serial cards with up to 32 terminals
attached to a PC running SCO unix.
I never touched SCO Unix, I got my Unix start on SunOS/Solaris and
very early Linux.
They were essentially IO processors
Sure, they have to be to handle than many serial lines.
But even in the "PC server" space, those were not that common.
Depends on your age and experience. I saw many. A multiuser system
that could handle 32 screens running a database oriented program for
< $10,000 was a very good alternative to a VAX.
Some of those setups persisted a surprisingly long time in the field.
Used to work in the back room at a call center that was still running a terminal-based setup 'til early 2012; when all you need is a form for
the reps to punch orders into, it Just Works.
(Never got into the backend enough to learn what the host was running;
some flavor of *nix that also managed the PBX, but the business end had
been built on a Win2k Server box in Visual FoxPro, of all things.)
Went into a bank in the mid noughties and the PC was running some sort
of remote terminal APP into a mainframe somewhere
But terminals are interesting. I read a bit on the documentation of
the VT52, where they talked a lot about microcode. The VT100 had a microprocessor already, an 8080.
Which is why it would have been trivial for DEC to build a reasonable
PC when they were building terminals.
The Natural Philosopher <tnp@invalid.invalid> schrieb:
Went into a bank in the mid noughties and the PC was running some sort
of remote terminal APP into a mainframe somewhere
The colored variant of the 3270 terminals or terminal emulators
(distinctive coloring easy to recognize) are in use at a furniture
store a few km from where I live. No idea what system they have
in the background, though.
But terminals are interesting. I read a bit on the documentation of the VT52, where they talked a lot about microcode. The VT100 had a
microprocessor already, an 8080.
Which is why it would have been trivial for DEC to build a reasonable
PC when they were building terminals.
A touch typist who can tab through fields is the fastest way to enter
data, and any eye candy is mere distraction
With rising fuel costs I somehow doubt supersonic airliners would ever
be profitable again
On 11/11/2026 09:08, Nuno Silva wrote:
On 2026-11-11, Lawrence DrCOOliveiro wrote:
On Tue, 10 Nov 2026 12:19:31 -0700, Peter Flass wrote:
Anyone would think that newer=faster. Who would have thought it
would be slower?
As another example, I recall it said that passenger airliners
after the Boeing 747 actually have slightly lower cruising speeds.
I guess itrCOs all part of the goal of saving fuel.
It is possible if the 747 was inefficient. IIRC there's a sweet spot
(or range?) for efficient fuel usage in subsonic jets.
707 was not a turbofan. As bypass engines became widely used, the
overall economics shifted to lower-a speed lower drag.
Airline economics is complex with passenger miles per day, and
maintenance costs per flight hour being balanced by fuel costs per
passenger mile.
Higher fuel costs as well as rising bypass ratios have increased the
tendency towards lower speed cruising
(For supersonic, IIRC Aerospatiale-BAC Concorde was actually quiteWith rising fuel costs I somehow doubt supersonic airliners would ever
efficient in cruise (compared to other supersonic aircraft), and I
think the proposed redesign would even do away with the need for wet
thrust (necessary for TO and transonic climb?).)
be profitable again
ISAM channel programs would look up the desired key in the index,
read the index record and store the disk address further down in the
channel program (I'm a bit vague on details, so I don't know if this
was done by the channel program or if there was a "program
controlled interrupt" (PCI) bit set to have an interrupt routine do
this), and then continue on and read the record - all in one I/O
operation.
I don't think supersonic airlines ever were profitable.
But terminals are interesting. I read a bit on the documentation of
the VT52, where they talked a lot about microcode. The VT100 had a microprocessor already, an 8080.
Which is why it would have been trivial for DEC to build a
reasonable PC when they were building terminals.
On 11/11/2026 16:42, John Ames wrote:
On Wed, 11 Nov 2026 12:36:43 +0000Went into a bank in the mid noughties and the PC was running some sort
The Natural Philosopher <tnp@invalid.invalid> wrote:
I see you never used the serial cards with up to 32 terminals
attached to a PC running SCO unix.
I never touched SCO Unix, I got my Unix start on SunOS/Solaris and
very early Linux.
They were essentially IO processors
Sure, they have to be to handle than many serial lines.
But even in the "PC server" space, those were not that common.
Depends on your age and experience. I saw many. A multiuser system
that could handle 32 screens running a database oriented program for
< $10,000 was a very good alternative to a VAX.
Some of those setups persisted a surprisingly long time in the field.
Used to work in the back room at a call center that was still running a
terminal-based setup 'til early 2012; when all you need is a form for
the reps to punch orders into, it Just Works.
(Never got into the backend enough to learn what the host was running;
some flavor of *nix that also managed the PBX, but the business end had
been built on a Win2k Server box in Visual FoxPro, of all things.)
of remote terminal APP into a mainframe somewhere
A touch typist who can tab through fields is the fastest way to enter
data, and any eye candy is mere distraction
Alexander Schreiber <als@usenet.thangorodrim.de> writes:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 10:37:49 +0100, Alexander Schreiber wrote:
... a key design aspect of mainframes back then was that I/O was not
done by the CPU (like it was done on home computers were all I/O was
entirely handled by the CPU), but fully off- loaded to peripherials
that were essentially limited computers of their own ...
Which made sense back when the CPU was a scarce and expensive resource.
It no longer made sense when that was no longer true.
It still makes sense and that is borne out in hardware design. We haven't >>flipped back to polled I/O because "CPUs are faster than we know what to
do with", quite the opposite: modern GPUs offload most the fancy 3D graphics >>processing (originally, these days it's also a lot of matrix math for AI), >>modern NICs offer a lot of offload (modern DC NICs are full on computers >>running Linux and e.g. handle network traffic encryption on their own,
see e.g. NVidia Bluefield DPUs) and so on.
https://www.marvell.com/products/data-processing-units.html
These are all peripherals, either directly soldered on
the mainboard, or via PCI-Express cards slots.
The Natural Philosopher <tnp@invalid.invalid> wrote:
On 11/11/2026 16:42, John Ames wrote:
On Wed, 11 Nov 2026 12:36:43 +0000Went into a bank in the mid noughties and the PC was running some sort
The Natural Philosopher <tnp@invalid.invalid> wrote:
I see you never used the serial cards with up to 32 terminals
attached to a PC running SCO unix.
I never touched SCO Unix, I got my Unix start on SunOS/Solaris and
very early Linux.
They were essentially IO processors
Sure, they have to be to handle than many serial lines.
But even in the "PC server" space, those were not that common.
Depends on your age and experience. I saw many. A multiuser system
that could handle 32 screens running a database oriented program for
< $10,000 was a very good alternative to a VAX.
Some of those setups persisted a surprisingly long time in the field.
Used to work in the back room at a call center that was still running a
terminal-based setup 'til early 2012; when all you need is a form for
the reps to punch orders into, it Just Works.
(Never got into the backend enough to learn what the host was running;
some flavor of *nix that also managed the PBX, but the business end had
been built on a Win2k Server box in Visual FoxPro, of all things.)
of remote terminal APP into a mainframe somewhere
Flying in the early 2010s, peeked at the display of the attendants
desktop at checkin: Some sort of Windows, but 90% of the screen was
occupied by a (IIRC) 3270 terminal emulator connected to the airline's
system and clearly showing a typical mainframe app.
As another example, I recall it said that passenger airliners after
the Boeing 747 actually have slightly lower cruising speeds. I guess
itrCOs all part of the goal of saving fuel.
It is possible if the 747 was inefficient. IIRC there's a sweet spot (or >range?) for efficient fuel usage in subsonic jets.
Alexander Schreiber <als@usenet.thangorodrim.de> writes:
The Natural Philosopher <tnp@invalid.invalid> wrote:
On 11/11/2026 16:42, John Ames wrote:
On Wed, 11 Nov 2026 12:36:43 +0000Went into a bank in the mid noughties and the PC was running some sort
The Natural Philosopher <tnp@invalid.invalid> wrote:
I see you never used the serial cards with up to 32 terminals
attached to a PC running SCO unix.
I never touched SCO Unix, I got my Unix start on SunOS/Solaris and >>>>>> very early Linux.
They were essentially IO processors
Sure, they have to be to handle than many serial lines.
But even in the "PC server" space, those were not that common.
Depends on your age and experience. I saw many. A multiuser system
that could handle 32 screens running a database oriented program for >>>>> < $10,000 was a very good alternative to a VAX.
Some of those setups persisted a surprisingly long time in the field.
Used to work in the back room at a call center that was still running a >>>> terminal-based setup 'til early 2012; when all you need is a form for
the reps to punch orders into, it Just Works.
(Never got into the backend enough to learn what the host was running; >>>> some flavor of *nix that also managed the PBX, but the business end had >>>> been built on a Win2k Server box in Visual FoxPro, of all things.)
of remote terminal APP into a mainframe somewhere
Flying in the early 2010s, peeked at the display of the attendants
desktop at checkin: Some sort of Windows, but 90% of the screen was >>occupied by a (IIRC) 3270 terminal emulator connected to the airline's >>system and clearly showing a typical mainframe app.
It's entirely likely that was communicating with a Unisys
2200/Clearpath system (UNIVAC Standard Airline System).
https://en.wikipedia.org/wiki/USAS_(application)
On 2026-11-11, The Natural Philosopher <tnp@invalid.invalid> wrote:
With rising fuel costs I somehow doubt supersonic airliners would ever
be profitable again
Besides, no matter how fast the plane flies you're still
going to be waiting for your baggage at the other end.
Actual air time is only part of the total travel time -
once you add in the trip to the airport, check-in time,
baggage retrieval, and transport from the destination
terminal to wherever you're really going, air time becomes
a relatively small portion of the total time.
Actual air time is only part of the total travel time -
once you add in the trip to the airport, check-in time,
baggage retrieval, and transport from the destination terminal to
wherever you're really going, air time becomes a relatively small
portion of the total time.
Went into a bank in the mid noughties and the PC was running some sortBy that time, I suspect the interface protocol would have been TN3270
of remote terminal APP into a mainframe somewhere
A touch typist who can tab through fields is the fastest way to enter
data, and any eye candy is mere distraction
The Natural Philosopher <tnp@invalid.invalid> schrieb:Indeed they did - The Rainbow PC-100 from 1982 was a dual-CPU version
Went into a bank in the mid noughties and the PC was running some sort
of remote terminal APP into a mainframe somewhere
The colored variant of the 3270 terminals or terminal emulators
(distinctive coloring easy to recognize) are in use at a furniture
store a few km from where I live. No idea what system they have
in the background, though.
But terminals are interesting. I read a bit on the documentation of the VT52, where they talked a lot about microcode. The VT100 had a
microprocessor already, an 8080.
Which is why it would have been trivial for DEC to build a reasonable
PC when they were building terminals.
The Natural Philosopher <tnp@invalid.invalid> wrote:
A touch typist who can tab through fields is the fastest way to enter
data, and any eye candy is mere distraction
Indeed. Just the learning curve is a bit steeper than most people these
days are used to.
The later VT200 series were nice and sleek and all that, but I felt
they were a step backwards in one respect: press the rCLSET-UPrCY key, and
it took perceptibly longer to switch to the setup menu than the old
VT100 did -- that was instant. I assume the older model had some double-buffering display hardware support which was dropped from the
newer range ...
The two engine 777 had more advanced and reliable engines. Once the 777
was ceritfied for long trips over water (known as ETOPS) the 747 was
dead. The 777 carried nearly as many people but with two engines and
a two person flight crew.
On 2026-11-11, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
The later VT200 series were nice and sleek and all that, but I felt
they were a step backwards in one respect: press the rCLSET-UPrCY key, and >> it took perceptibly longer to switch to the setup menu than the old
VT100 did -- that was instant. I assume the older model had some
double-buffering display hardware support which was dropped from the
newer range ...
It was a step backwards in another respect too: it indulged
in the same sort of keyboard layout shenanigans as the IBM PC.
I refused to touch one for that reason.
It took the industry at least 10 years to recover from the
silliness that the IBM PC keyboard layout inflicted on it.
On 2026-11-11, John Levine <johnl@taugh.com> wrote:
The two engine 777 had more advanced and reliable engines. Once the 777
was ceritfied for long trips over water (known as ETOPS) the 747 was
dead. The 777 carried nearly as many people but with two engines and
a two person flight crew.
ETOPS officially stood for "Extended Twin-engine OPerationS",
but it was colloquially known as "Engines Turn Or Passengers Swim".
On 2026-11-11, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
The later VT200 series were nice and sleek and all that, but I felt
they were a step backwards in one respect: press the rCLSET-UPrCY key,
and it took perceptibly longer to switch to the setup menu than the
old VT100 did -- that was instant. I assume the older model had
some double-buffering display hardware support which was dropped
from the newer range ...
It was a step backwards in another respect too: it indulged in the
same sort of keyboard layout shenanigans as the IBM PC.
I refused to touch one for that reason.
An engineer would say that with 896 KB of RAM [the DEC Rainbow] was
superior to the IBM-PC; but it did not have an ISA bus, so it would
never be clone-level compatible.
But in my opinion, the DecMate based on a single-chip PDP-8 was a much superior PC for office use.
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
.
On Tue, 10 Nov 2026 10:37:49 +0100, Alexander Schreiber wrote:
... a key design aspect of mainframes back then was that I/O was
not done by the CPU (like it was done on home computers were all
I/O was entirely handled by the CPU), but fully off- loaded to
peripherials that were essentially limited computers of their own
...
Which made sense back when the CPU was a scarce and expensive
resource. It no longer made sense when that was no longer true.
It still makes sense and that is borne out in hardware design. We
haven't flipped back to polled I/O because "CPUs are faster than we
know what to do with", quite the opposite: modern GPUs offload most
the fancy 3D graphics processing (originally, these days it's also a
lot of matrix math for AI), modern NICs offer a lot of offload
(modern DC NICs are full on computers running Linux and e.g. handle
network traffic encryption on their own, see e.g. NVidia Bluefield
DPUs) and so on.
Nowadays we have millions of users on social media simultaneously,
but that request/response paradigm is no longer sufficient for the
real-time interaction requirements.
The "Unix serial terminal" model never applied to web interactive
traffic, that was always running on the "smart terminal doing local
edits - mainframe style" model.
Computing HARDWARE has gotten so much cheaper it makes sense to put
some intelligence into any device so "the computer" can just ask it
to do something and go do something else.
Friends of mine are playing with those a lot, first one of them
managed to run DOOM on it and the latest trick was connecting a
BlueField 2 via PCIe interface to a regular (if somewhat older) PC
GPU and having the Bluefield drive _that_ to run graphics. No
"regular" PC required at all.
On Wed, 11 Nov 2026 07:45:23 -0700, Peter Flass wrote:
Computing HARDWARE has gotten so much cheaper it makes sense to put
some intelligence into any device so "the computer" can just ask it
to do something and go do something else.
Or, to put it another way, multiple CPU cores have become so
commonplace that it is no longer considered a big deal to rCLwasterCY some
of them busy-waiting without bogging down others that are doing actual
useful work.
Indeed they did - The Rainbow PC-100 from 1982 was a dual-CPU version
with a VT220 like screen. It had a Z80 to run CP/M and an 8088 to run
MS-DOS. An engineer would say that with 896 KB of RAM it was superior to
the IBM-PC; but it did not have an ISA bus, so it would never be
clone-level compatible.
On 11/12/26 06:35, Lawrence DrCOOliveiro wrote:
On Wed, 11 Nov 2026 07:45:23 -0700, Peter Flass wrote:
Computing HARDWARE has gotten so much cheaper it makes sense to
put some intelligence into any device so "the computer" can just
ask it to do something and go do something else.
Or, to put it another way, multiple CPU cores have become so
commonplace that it is no longer considered a big deal to rCLwasterCY
some of them busy-waiting without bogging down others that are
doing actual useful work.
I read their comments to be the exact opposite, and, FWIW, it does
seem to me that event/interrupt driven software and devices have
become more prevalent.
I'm still having difficulty understanding the difference between a
mainframe and a PC.
But after Windows NT, I never understood the difference between a
mini computer and a PC.
Perhaps next we can do the difference between a SoC computer (which
I'm actually using now) and a PC ;-).
Actual air time is only part of the total travel time -
once you add in the trip to the airport, check-in time,
baggage retrieval, and transport from the destination
terminal to wherever you're really going, air time becomes
a relatively small portion of the total time.
But after Windows NT, I never understood the difference between a mini computer and a PC. Perhaps next we can do the difference between a SoC computer (which I'm actually using now) and a PC
On 2026-11-11, Alexander Schreiber <als@usenet.thangorodrim.de> wrote:
The Natural Philosopher <tnp@invalid.invalid> wrote:
A touch typist who can tab through fields is the fastest way to enter
data, and any eye candy is mere distraction
Indeed. Just the learning curve is a bit steeper than most people these
days are used to.
Perhaps, but oh, the advantages once you've learned it.
And you only had to learn it once, rather than having to
start all over again when each year's new model comes out.
On Wed, 11 Nov 2026 12:15:20 +0100, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
.
On Tue, 10 Nov 2026 10:37:49 +0100, Alexander Schreiber wrote:
... a key design aspect of mainframes back then was that I/O was
not done by the CPU (like it was done on home computers were all
I/O was entirely handled by the CPU), but fully off- loaded to
peripherials that were essentially limited computers of their own
...
Which made sense back when the CPU was a scarce and expensive
resource. It no longer made sense when that was no longer true.
It still makes sense and that is borne out in hardware design. We
haven't flipped back to polled I/O because "CPUs are faster than we
know what to do with", quite the opposite: modern GPUs offload most
the fancy 3D graphics processing (originally, these days it's also a
lot of matrix math for AI), modern NICs offer a lot of offload
(modern DC NICs are full on computers running Linux and e.g. handle
network traffic encryption on their own, see e.g. NVidia Bluefield
DPUs) and so on.
So really, you are saying that general-purpose CPUs are now offloading
work ... onto other general-purpose CPUs.
Nowadays we have millions of users on social media simultaneously,
but that request/response paradigm is no longer sufficient for the
real-time interaction requirements.
The "Unix serial terminal" model never applied to web interactive
traffic, that was always running on the "smart terminal doing local
edits - mainframe style" model.
Again, yourCOre thinking in terms of an analogue of the old mainframe block-mode terminal. Real-time interaction requirements on the Web
involve more than just the strict request/response model: you also
have AJAX and WebSockets to provide a far greater level of
interactivity. Block-mode terminals have no analogue to such
functionality.
On Wed, 11 Nov 2026 07:45:23 -0700, Peter Flass wrote:
Computing HARDWARE has gotten so much cheaper it makes sense to put
some intelligence into any device so "the computer" can just ask it
to do something and go do something else.
Or, to put it another way, multiple CPU cores have become so
commonplace that it is no longer considered a big deal to rCLwasterCY some
of them busy-waiting without bogging down others that are doing actual
useful work.
On 11/11/2026 12:33, Charlie Gibbs wrote:
On 2026-11-11, The Natural Philosopher <tnp@invalid.invalid> wrote:
With rising fuel costs I somehow doubt supersonic airliners would ever
be profitable again
Besides, no matter how fast the plane flies you're still
going to be waiting for your baggage at the other end.
Actual air time is only part of the total travel time -
once you add in the trip to the airport, check-in time,
baggage retrieval, and transport from the destination
terminal to wherever you're really going, air time becomes
a relatively small portion of the total time.
On a transoceanic route, time in the air is quite significant.
On a short "puddlejumper" route, a high-speed train (TGV in French) may
well be faster.
On 11/11/2026 12:33, Charlie Gibbs wrote:
On 2026-11-11, The Natural Philosopher <tnp@invalid.invalid> wrote:
With rising fuel costs I somehow doubt supersonic airliners would ever
be profitable again
Besides, no matter how fast the plane flies you're still
going to be waiting for your baggage at the other end.
Actual air time is only part of the total travel time -
once you add in the trip to the airport, check-in time,
baggage retrieval, and transport from the destination
terminal to wherever you're really going, air time becomes
a relatively small portion of the total time.
On a transoceanic route, time in the air is quite significant.
On a short "puddlejumper" route, a high-speed train (TGV in French)
may well be faster.
c186282 <c186282@nnada.net> wrote:
On 11/9/26 16:20, John Levine wrote:
According to c186282 <c186282@nnada.net>:
Noted in this group though - I/O arch diffs between
mainframes and minis.
Will someone try to elaborate ?
The classic difference is that mainframes have channels, so the
processor sets up a bunch of work for the channel to do, the channel
goes off and does it independently of the CPU. Minis had much less
complex I/O controllers so they do more detailed stuff. An example I
gave a few messages back was for a card reader. A mainframe starts a
channel program, it reads in a card or maybe seveal cards, and reports
back with an interrupt. On a mini there was an I/O operation to start
the card reader, an interrupt for each column and an I/O to read it,
and then something at the end to stop the reader or go to the next
card.
But that was 40 or 50 years ago. These days even cheap generic disk
controllers do more than channels did in the 1960s, e.g., keep a queue
of requests and schedule them to minimize delay, and handle spare block
replacement for bad blocks.
Modern mainframes have extremely capable channels but the distinction is >>> more about reliability and things like hardware and software upgrades in >>> place without rebooting.
Ok ... so 'mainframes' as you saw them were more
'directors' for distributed processing rather
than THE processor.
No. The operating system and application software ran on the mainframe
CPU. But - as described above - a key design aspect of mainframes back
then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully off- loaded to peripherials that were essentially limited computers of their
own:
On 2026-11-11, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
The later VT200 series were nice and sleek and all that, but I felt
they were a step backwards in one respect: press the rCLSET-UPrCY key, and >> it took perceptibly longer to switch to the setup menu than the old
VT100 did -- that was instant. I assume the older model had some
double-buffering display hardware support which was dropped from the
newer range ...
It was a step backwards in another respect too: it indulged
in the same sort of keyboard layout shenanigans as the IBM PC.
I refused to touch one for that reason.
It took the industry at least 10 years to recover from the
silliness that the IBM PC keyboard layout inflicted on it.
On Wed, 11 Nov 2026 12:15:20 +0100, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
.
On Tue, 10 Nov 2026 10:37:49 +0100, Alexander Schreiber wrote:
... a key design aspect of mainframes back then was that I/O was
not done by the CPU (like it was done on home computers were all
I/O was entirely handled by the CPU), but fully off- loaded to
peripherials that were essentially limited computers of their own
...
Which made sense back when the CPU was a scarce and expensive
resource. It no longer made sense when that was no longer true.
It still makes sense and that is borne out in hardware design. We
haven't flipped back to polled I/O because "CPUs are faster than we
know what to do with", quite the opposite: modern GPUs offload most
the fancy 3D graphics processing (originally, these days it's also a
lot of matrix math for AI), modern NICs offer a lot of offload
(modern DC NICs are full on computers running Linux and e.g. handle
network traffic encryption on their own, see e.g. NVidia Bluefield
DPUs) and so on.
So really, you are saying that general-purpose CPUs are now offloading
work ... onto other general-purpose CPUs.
On 11/12/26 06:35, Lawrence DrCOOliveiro wrote:
On Wed, 11 Nov 2026 07:45:23 -0700, Peter Flass wrote:
Computing HARDWARE has gotten so much cheaper it makes sense to put
some intelligence into any device so "the computer" can just ask it
to do something and go do something else.
Or, to put it another way, multiple CPU cores have become so
commonplace that it is no longer considered a big deal to rCLwasterCY some >> of them busy-waiting without bogging down others that are doing actual
useful work.
I read their comments to be the exact opposite, and, FWIW, it does seem
to me that event/interrupt driven software and devices have become more prevalent.
I'm still having difficulty understanding the difference between a
mainframe and a PC. The argument that mainframes have independent IO
devices appears to become "more" independent IO devices, when it is
pointed out PCs have independent IO devices. It sounds a bit like an argument that mainframes are mainframes because they use electricity,
and when it is pointed out PCs also use electricity the argument becomes there is no comparison because mainframes use kilowatts and PCs only use
a few watts. Everything is true, but the distinctions are blurred.
On Wed, 11 Nov 2026 19:03:01 -0800, Lars Poulsen wrote:
Indeed they did - The Rainbow PC-100 from 1982 was a dual-CPU version
with a VT220 like screen. It had a Z80 to run CP/M and an 8088 to run
MS-DOS. An engineer would say that with 896 KB of RAM it was superior to
the IBM-PC; but it did not have an ISA bus, so it would never be
clone-level compatible.
Another problem is DEC attempted to capture the diskette sales with a proprietary format available only from them at a much higher cost than generic Elephant or Memorex diskettes. DEC was still living in the mini world.
On 11/12/26 00:28, rbowman wrote:
On Wed, 11 Nov 2026 19:03:01 -0800, Lars Poulsen wrote:
Indeed they did - The Rainbow PC-100 from 1982 was a dual-CPU version
with a VT220 like screen. It had a Z80 to run CP/M and an 8088 to run
MS-DOS. An engineer would say that with 896 KB of RAM it was superior to >>> the IBM-PC; but it did not have an ISA bus, so it would never be
clone-level compatible.
Another problem is DEC attempted to capture the diskette sales with a
proprietary format available only from them at a much higher cost than
generic Elephant or Memorex diskettes. DEC was still living in the mini
world.
DEC had so many offerings people can be forgiven for not keeping them >straight. They had Rainbow and Robin (PCs, I think) and VaxStation and >DECStation. Anything else?
On 2026-11-11, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
The later VT200 series were nice and sleek and all that, but I felt
they were a step backwards in one respect: press the rCLSET-UPrCY key, and >> it took perceptibly longer to switch to the setup menu than the old
VT100 did -- that was instant. I assume the older model had some
double-buffering display hardware support which was dropped from the
newer range ...
It was a step backwards in another respect too: it indulged
in the same sort of keyboard layout shenanigans as the IBM PC.
I refused to touch one for that reason.
It took the industry at least 10 years to recover from the
silliness that the IBM PC keyboard layout inflicted on it.
On 11/11/2026 20:33, Charlie Gibbs wrote:
Actual air time is only part of the total travel time -
once you add in the trip to the airport, check-in time,
baggage retrieval, and transport from the destination
terminal to wherever you're really going, air time becomes
a relatively small portion of the total time.
Not if you are flying to Australia from the UK
But yes, the break-even with driving is around 400 miles - less if
there is congestion or a ferry or ocean in between.
f you're going to develop an intelligent peripheral or controller, it
makes both economic and technical sense to grab an off-the-shelf general purpose CPU. Postscript printers were an early example of this in the PC world. I think there were some systems where the printer had more
compute power than the using system.
The Natural Philosopher <tnp@invalid.invalid> writes:
On 11/11/2026 20:33, Charlie Gibbs wrote:
Actual air time is only part of the total travel time -
once you add in the trip to the airport, check-in time,
baggage retrieval, and transport from the destination
terminal to wherever you're really going, air time becomes
a relatively small portion of the total time.
Not if you are flying to Australia from the UK
But yes, the break-even with driving is around 400 miles - less if
there is congestion or a ferry or ocean in between.
I dunno. I used to drive from San Jose to Pasadena (350 miles). The
trip took 4 hours 50 minutes[*]. Flight time from San Jose to Burbank
is less than an hour (and before 2001, one could arrive at the
gate 10 minutes before departure) and the rental cars were right
there at the terminal, a five minute walk from baggage claim.
[*] Assuming no significant traffic blockages and no speeding tickets
(before 1987, the speed limit on I5 was 55mph - raised from 65 to 70 in 1996).
On 11/11/26 20:44, Charlie Gibbs wrote:
On 2026-11-11, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
The later VT200 series were nice and sleek and all that, but I felt
they were a step backwards in one respect: press the rCLSET-UPrCY key, and >>> it took perceptibly longer to switch to the setup menu than the old
VT100 did -- that was instant. I assume the older model had some
double-buffering display hardware support which was dropped from the
newer range ...
It was a step backwards in another respect too: it indulged
in the same sort of keyboard layout shenanigans as the IBM PC.
I refused to touch one for that reason.
It took the industry at least 10 years to recover from the
silliness that the IBM PC keyboard layout inflicted on it.
1. I loved the VT220.
2. The keyboard I'm typing this on a Unicomp a PC-101 clone.
I refuse to touch a keyboard with an abomination called a "Windows
Key" on it.
In fact if you go back to the TMS9900 series of chips these made
lousy single user computers. The chipset was designed for efficient
context switching in a multitasking environment
On Wed, 11 Nov 2026 19:03:01 -0800, Lars Poulsen wrote:
Indeed they did - The Rainbow PC-100 from 1982 was a dual-CPU version
with a VT220 like screen. It had a Z80 to run CP/M and an 8088 to run
MS-DOS. An engineer would say that with 896 KB of RAM it was superior to
the IBM-PC; but it did not have an ISA bus, so it would never be
clone-level compatible.
Another problem is DEC attempted to capture the diskette sales with a proprietary format available only from them at a much higher cost than generic Elephant or Memorex diskettes. DEC was still living in the mini world.
Charlie Gibbs <cgibbs@kltpzyxm.invalid> writes:
It took the industry at least 10 years to recover from the
silliness that the IBM PC keyboard layout inflicted on it.
It hasn't really recovered - the caps lock and control
keys are still in the wrong place on PC keyboards. Fortunately,
that's easy to fix on Unix/Linux and MacOs systems.
I don't use windows (never have), so I can't speak for it.
On 11/11/26 22:44, Charlie Gibbs wrote:
On 2026-11-11, John Levine <johnl@taugh.com> wrote:
The two engine 777 had more advanced and reliable engines. Once the 777 >>> was ceritfied for long trips over water (known as ETOPS) the 747 was
dead. The 777 carried nearly as many people but with two engines and
a two person flight crew.
ETOPS officially stood for "Extended Twin-engine OPerationS",
but it was colloquially known as "Engines Turn Or Passengers Swim".
Heh heh !
Frankly I'd be MUCH happier with at least THREE
engines if going over wide expanses.
On Thu, 12 Nov 2026 03:44:53 GMT, Charlie Gibbs wrote:
On 2026-11-11, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
The later VT200 series were nice and sleek and all that, but I felt
they were a step backwards in one respect: press the rCLSET-UPrCY key,
and it took perceptibly longer to switch to the setup menu than the
old VT100 did -- that was instant. I assume the older model had
some double-buffering display hardware support which was dropped
from the newer range ...
It was a step backwards in another respect too: it indulged in the
same sort of keyboard layout shenanigans as the IBM PC.
No it didnrCOt. Compare <http://xahlee.info/kbd/vt220_terminal.html>.
I refused to touch one for that reason.
The short travel was unusual for the time, and was recognized as
allowing for faster typing, but could also aggravate RSI.
Mainframes are optimised for extremely large numbers of users. Probably using intelligent terminals that send a batch of data after many
keystrokes. Attached to extremely large databases. On very resilient hardware.
But yes, the break-even with driving is around 400 miles - less if
there is congestion or a ferry or ocean in between.
London to New York would be about 3 days by fast nuclear powered ship.
On 2026-11-12, The Natural Philosopher <tnp@invalid.invalid> wrote:
Mainframes are optimised for extremely large numbers of users. Probably
using intelligent terminals that send a batch of data after many
keystrokes. Attached to extremely large databases. On very resilient
hardware.
OK, let's take another stab at it, starting with your word "batch". >Mainframes are traditionally batch-oriented. You submit jobs and
data files, and it chews through them and leaves you the results.
Interaction with the user is limited and often nonexistent.
In fact if you go back to the TMS9900 series of chips these made lousy
single user computers. The chipset was designed for efficient context switching in a multitasking environment
2. The keyboard I'm typing this on a Unicomp a PC-101 clone.
I refuse to touch a keyboard with an abomination called a "Windows
Key" on it.
According to The Natural Philosopher <tnp@invalid.invalid>:
But yes, the break-even with driving is around 400 miles - less if
there is congestion or a ferry or ocean in between.
London to New York would be about 3 days by fast nuclear powered ship.
The SS United States went frm New York to England in 3 days, 10 hours, 40 minutes.
That was in 1952.
The ship was recently in the news having been towed to Florida and sunk yesterday
to make an artificial reef for divers.
R's,--
John
On 11/11/26 23:32, Lawrence DrCOOliveiro wrote:
If you're going to develop an intelligent peripheral or controller, it
So really, you are saying that general-purpose CPUs are now
offloading work ... onto other general-purpose CPUs.
makes both economic and technical sense to grab an off-the-shelf general purpose CPU.
I refuse to touch a keyboard with an abomination called a "Windows
Key" on it.
On 2026-11-12, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Thu, 12 Nov 2026 03:44:53 GMT, Charlie Gibbs wrote:
It was a step backwards in another respect too: it indulged in the
same sort of keyboard layout shenanigans as the IBM PC.
No it didnrCOt. Compare <http://xahlee.info/kbd/vt220_terminal.html>.
I see the same IBM tricks there: the extra key between Z and the
left-hand shift key, and the return key moved farther away.
The Windows key can be made useful by remapping it to Compose.
On 2026-11-12, c186282 <c186282@nnada.net> wrote:
On 11/11/26 22:44, Charlie Gibbs wrote:
On 2026-11-11, John Levine <johnl@taugh.com> wrote:
The two engine 777 had more advanced and reliable engines. Once the 777 >>>> was ceritfied for long trips over water (known as ETOPS) the 747 was
dead. The 777 carried nearly as many people but with two engines and
a two person flight crew.
ETOPS officially stood for "Extended Twin-engine OPerationS",
but it was colloquially known as "Engines Turn Or Passengers Swim".
Heh heh !
Frankly I'd be MUCH happier with at least THREE
engines if going over wide expanses.
That's why ETOPS was devised. It contains measures to improve
reliability. For instance, a different crew will work on each
engine, to prevent one group's possible systematic errors from
affecting the other.
Still, shit happens. The DC-10 in the Sioux City crash had
three separate hydraulic systems - and the disintegrating
engine took out all three. BTW that plane did have three
engines.
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
On Thu, 12 Nov 2026 07:32:20 -0700, Peter Flass wrote:
I refuse to touch a keyboard with an abomination called a "Windows
Key" on it.
My Das Keyboard 4 has an interesting wing-like symbol on that key.
Other hardware (like laptops which didnrCOt come with Windows) uses
other symbols -- my System76 laptop labels it the rCLSuperrCY key, in the true hacker tradition. I think I have seen other Linux laptops that
put a penguin symbol there.
Microsoft even tried adding another one lately, the rCLCopilotrCY key. I donrCOt think thatrCOs been a big success ...
Most important IO on home computers was to video screen. Most home
computers did that in hardware. Some had extra processing, like
sprites on Commodore 64.
On 11/12/26 13:43, Lawrence DrCOOliveiro wrote:
On Thu, 12 Nov 2026 07:32:20 -0700, Peter Flass wrote:
I refuse to touch a keyboard with an abomination called a "Windows
Key" on it.
My Das Keyboard 4 has an interesting wing-like symbol on that key.
Other hardware (like laptops which didnrCOt come with Windows) uses
other symbols -- my System76 laptop labels it the rCLSuperrCY key, in the
true hacker tradition. I think I have seen other Linux laptops that
put a penguin symbol there.
Microsoft even tried adding another one lately, the rCLCopilotrCY key. I
donrCOt think thatrCOs been a big success ...
That's because God is my copilot ;-)
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous failures.
On Thu, 12 Nov 2026 18:46:30 GMT, Charlie Gibbs wrote:
On 2026-11-12, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Thu, 12 Nov 2026 03:44:53 GMT, Charlie Gibbs wrote:
It was a step backwards in another respect too: it indulged in the
same sort of keyboard layout shenanigans as the IBM PC.
No it didnrCOt. Compare <http://xahlee.info/kbd/vt220_terminal.html>.
I see the same IBM tricks there: the extra key between Z and the
left-hand shift key, and the return key moved farther away.
You mean, the angle brackets moved to their own key, so that rCL,rCY and rCL.rCY are still accessible with or without shift?
In some ways, I think that makes sense. I still regularly hit rCL,rCY or rCL.rCY without letting go of the shift key, and get rCL<rCY and rCL>rCY instead
without meaning to. Having dedicated keys would have been another way
that typists could work even faster.
The Amiga had the cleverest video hardware of this era, in the form of
the rCLcopperrCY chip. Instead of a simple RAM buffer for video, this implemented a display list, where each scan line of the display could
come from a different block of memory. This allowed for all kinds of
clever tricks by fiddling list pointers while the screen was still
being redrawn; these were fast because you didnrCOt need to copy large
blocks of memory around.
On 11/11/2026 11:03, Alexander Schreiber wrote:
The Natural Philosopher <tnp@invalid.invalid> wrote:Depends on your age and experience. I saw many. A multiuser system that could handle 32 screens running a database oriented program for <
On 10/11/2026 09:37, Alexander Schreiber wrote:
I/O was not done by the CPU (like it was done on homeI see you never used the serial cards with up to 32 terminals attached
computers were all I/O was entirely handled by the CPU), but fully off- >>>> loaded to peripherials that were essentially limited computers of their >>>> own:
to a PC running SCO unix.
I never touched SCO Unix, I got my Unix start on SunOS/Solaris and very
early Linux.
They were essentially IO processors
Sure, they have to be to handle than many serial lines.
But even in the "PC server" space, those were not that common.
$10,000 was a very good alternative to a VAX.
Adding in TCP/IP rendered the serial ports obsolete and made it even
better, apart from the cost per desktop of a PC running
telnet...compared to a VT220 or a Wyse 50.
On 2026-11-12, Alexander Schreiber <als@usenet.thangorodrim.de> wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous failures.
That's one of the things I like about aviation: it's one of
the few remaining places where safety is taken seriously.
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
On 2026-11-12, c186282 <c186282@nnada.net> wrote:
On 11/11/26 22:44, Charlie Gibbs wrote:
On 2026-11-11, John Levine <johnl@taugh.com> wrote:
The two engine 777 had more advanced and reliable engines. Once the 777 >>>>> was ceritfied for long trips over water (known as ETOPS) the 747 was >>>>> dead. The 777 carried nearly as many people but with two engines and >>>>> a two person flight crew.
ETOPS officially stood for "Extended Twin-engine OPerationS",
but it was colloquially known as "Engines Turn Or Passengers Swim".
Heh heh !
Frankly I'd be MUCH happier with at least THREE
engines if going over wide expanses.
That's why ETOPS was devised. It contains measures to improve
reliability. For instance, a different crew will work on each
engine, to prevent one group's possible systematic errors from
affecting the other.
And we have seen accidents that root-caused to "mechanic messed up
during maintenance", e.g. the SmartLynx Flight 9001 Incident (hard
landing after loss of control, root-caused to wrong lubricant applied).
Still, shit happens. The DC-10 in the Sioux City crash had
three separate hydraulic systems - and the disintegrating
engine took out all three. BTW that plane did have three
engines.
And that day it was learned that "three independent things" is not
enough, their components also need to be set up not to share failure
risks - such as being torn apart by a disintegrating engine near
all three hydraulic lines. Sadly, this lesson did cost lives.
--Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous failures.
Kind regards,
Alex.
On 2026-11-12, Peter Flass <Peter@Iron-Spring.com> wrote:
On 11/12/26 13:43, Lawrence DrCOOliveiro wrote:
On Thu, 12 Nov 2026 07:32:20 -0700, Peter Flass wrote:
I refuse to touch a keyboard with an abomination called a "Windows
Key" on it.
My Das Keyboard 4 has an interesting wing-like symbol on that key.
Other hardware (like laptops which didnrCOt come with Windows) uses
other symbols -- my System76 laptop labels it the rCLSuperrCY key, in the >>> true hacker tradition. I think I have seen other Linux laptops that
put a penguin symbol there.
Microsoft even tried adding another one lately, the rCLCopilotrCY key. I >>> donrCOt think thatrCOs been a big success ...
That's because God is my copilot ;-)
That's better than Copilot being your god, I guess.
On Thu, 12 Nov 2026 10:33:48 -0000 (UTC), Waldek Hebisch wrote:
Most important IO on home computers was to video screen. Most home
computers did that in hardware. Some had extra processing, like
sprites on Commodore 64.
An early Sinclair machine (ZX80? ZX81?) cut costs by having the screen refresh managed by the CPU in place of dedicated hardware. Every time
you pressed a key on the keyboard, the screen would go blank for a
fraction of a second, while the CPU stopped screen refresh to process
your input.
Distracting, yes, but then, what else could you get for that price?
The early Apple Mac models didnrCOt have dedicated VRAM. Instead, part
of main DRAM was used for the screen display and the sound buffers.
The hardware for refreshing these stole cycles from the CPU to do its updates; thus, the nominal CPU speed of 7.8336MHz was effectively
reduced by about 1-2MHz as a result of this memory contention.
The Amiga had the cleverest video hardware of this era, in the form of--
the rCLcopperrCY chip. Instead of a simple RAM buffer for video, this implemented a display list, where each scan line of the display could
come from a different block of memory. This allowed for all kinds of
clever tricks by fiddling list pointers while the screen was still
being redrawn; these were fast because you didnrCOt need to copy large
blocks of memory around.
"Copper" was short for "co-processor". Yup, you could say it was a
form of channel. The Amiga also had a boatload of DMA channels.
Omigod, there goes the line between PC and mainframe again...
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Wed, 11 Nov 2026 12:15:20 +0100, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
.
Nowadays we have millions of users on social media
simultaneously, but that request/response paradigm is no longer
sufficient for the real-time interaction requirements.
The "Unix serial terminal" model never applied to web interactive
traffic, that was always running on the "smart terminal doing
local edits - mainframe style" model.
Again, yourCOre thinking in terms of an analogue of the old mainframe
block-mode terminal. Real-time interaction requirements on the Web
involve more than just the strict request/response model: you also
have AJAX and WebSockets to provide a far greater level of
interactivity. Block-mode terminals have no analogue to such
functionality.
You are thinking too much from a "what the user sees" perspective.
Under the hood all the "Real-time interaction requirements on the
Web" boil down to RPC traffic - be it a browser loading a page or
the JavaScript running on that page fetching more data.
In fact one can think of the mainframe block mode terminals as RPC
clients ...
On Thu, 12 Nov 2026 22:41:03 GMT, Charlie Gibbs wrote:
"Copper" was short for "co-processor". Yup, you could say it was a
form of channel. The Amiga also had a boatload of DMA channels.
Omigod, there goes the line between PC and mainframe again...
Did mainframes have channel controllers where you could dynamically
alter the program as the channel was running?
Somehow I donrCOt think they did -- there would have been no rCLbusinessrCY use for such functionality. You just set up the program, sent the
START command to the channel, and went away and did something else
until it told you it had finished.
On Thu, 12 Nov 2026 22:41:03 GMT, Charlie Gibbs wrote:
"Copper" was short for "co-processor". Yup, you could say it was a
form of channel. The Amiga also had a boatload of DMA channels.
Omigod, there goes the line between PC and mainframe again...
Did mainframes have channel controllers where you could dynamically
alter the program as the channel was running?
On 11/12/26 18:58, Lawrence DrCOOliveiro wrote:
On Thu, 12 Nov 2026 22:41:03 GMT, Charlie Gibbs wrote:
"Copper" was short for "co-processor". Yup, you could say it was a
form of channel. The Amiga also had a boatload of DMA channels.
Omigod, there goes the line between PC and mainframe again...
Did mainframes have channel controllers where you could dynamically
alter the program as the channel was running?
Somehow I donrCOt think they did -- there would have been no
rCLbusinessrCY use for such functionality. You just set up the program,
sent the START command to the channel, and went away and did
something else until it told you it had finished.
Yes, they did. I previously mentioned ISAM. Lynne could probably say
more about this, but I believe that for a long time OS/360 programs
using ISAM couldn't run in a virtual machine because of potential
timing problems.
On 2026-11-12, Alexander Schreiber <als@usenet.thangorodrim.de> wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous failures.
That's one of the things I like about aviation: it's one of
the few remaining places where safety is taken seriously.
So really, you are saying that general-purpose CPUs are now offloading
work ... onto other general-purpose CPUs.
On Thu, 12 Nov 2026 10:49:34 +0100, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Wed, 11 Nov 2026 12:15:20 +0100, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
.
Nowadays we have millions of users on social media
simultaneously, but that request/response paradigm is no longer
sufficient for the real-time interaction requirements.
The "Unix serial terminal" model never applied to web interactive
traffic, that was always running on the "smart terminal doing
local edits - mainframe style" model.
Again, yourCOre thinking in terms of an analogue of the old mainframe
block-mode terminal. Real-time interaction requirements on the Web
involve more than just the strict request/response model: you also
have AJAX and WebSockets to provide a far greater level of
interactivity. Block-mode terminals have no analogue to such
functionality.
You are thinking too much from a "what the user sees" perspective.
Perception doesnrCOt count. ThatrCOs reality. You canrCOt *pretend* to offer interactivity: you really have to have support for interactivity built
into your architecture from the lowest levels up. You canrCOt build per-character response on top of block mode.
Under the hood all the "Real-time interaction requirements on the
Web" boil down to RPC traffic - be it a browser loading a page or
the JavaScript running on that page fetching more data.
WebSocket connections are asynchronous full-duplex connections --
thatrCOs not rCLRPCrCY by any stretch.
Think of how rCLdoomscrollingrCY works: the page is continually fetching
more data to update itself.
Think of how Discord works: type some text and press Enter, and your
message is immediately visible to everyone else on the same channel.
Just start typing, and others are notified of that, too, so they can
wait to see what you are going to say before adding their own
comments.
In fact one can think of the mainframe block mode terminals as RPC
clients ...
Maybe you can, but thatrCOs not how the interactive Web works.
Just because most web app developers these days never dive down all the way to the bottom of their stack where the RPCs happen doesn't mean they aren't happening.Exactly. Browser with JavaScript = thick client = equivalent 3270 functionality
Kind regards,
Alex.
On Thu, 12 Nov 2026 20:00:03 -0700, Peter Flass wrote:
On 11/12/26 18:58, Lawrence DrCOOliveiro wrote:
On Thu, 12 Nov 2026 22:41:03 GMT, Charlie Gibbs wrote:
"Copper" was short for "co-processor". Yup, you could say it was a
form of channel. The Amiga also had a boatload of DMA channels.
Omigod, there goes the line between PC and mainframe again...
Did mainframes have channel controllers where you could dynamically
alter the program as the channel was running?
Somehow I donrCOt think they did -- there would have been no
rCLbusinessrCY use for such functionality. You just set up the program,
sent the START command to the channel, and went away and did
something else until it told you it had finished.
Yes, they did. I previously mentioned ISAM. Lynne could probably say
more about this, but I believe that for a long time OS/360 programs
using ISAM couldn't run in a virtual machine because of potential
timing problems.
That sounds more like a bug than a feature: it indicates that any
attempt to change the channel program while it was running just ended
up confusing the channel into misbehaving -- as opposed to redirecting
it gracefully onto a new path on the fly.
Remember, the Amiga was actually designed to cope with this: updates
to the display list were interlocked with the CPU to maintain
consistency.
Unlike your IBM mainframe, the Amiga designers had no choice but to do
it this way: the copper never stopped running, otherwise the user
would have no screen display to look at.
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-12, Alexander Schreiber <als@usenet.thangorodrim.de> wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous failures.
That's one of the things I like about aviation: it's one of
the few remaining places where safety is taken seriously.
I work in the chmemical industry, and we certainly take safety
seriously there.
But if you're interested in airline safety, I assume you read the
Aviation Herald? One disturbing trend is the increasing number
of lithium ion battery fires from things that passengers brought
on board.
On 11/12/26 15:41, Charlie Gibbs wrote:
On 2026-11-12, Peter Flass <Peter@Iron-Spring.com> wrote:https://en.wikipedia.org/wiki/God_Is_My_Co-Pilot_(film)
On 11/12/26 13:43, Lawrence DrCOOliveiro wrote:
On Thu, 12 Nov 2026 07:32:20 -0700, Peter Flass wrote:
I refuse to touch a keyboard with an abomination called a "Windows
Key" on it.
My Das Keyboard 4 has an interesting wing-like symbol on that key.
Other hardware (like laptops which didnrCOt come with Windows) uses
other symbols -- my System76 laptop labels it the rCLSuperrCY key, in the >>>> true hacker tradition. I think I have seen other Linux laptops that
put a penguin symbol there.
Microsoft even tried adding another one lately, the rCLCopilotrCY key. I >>>> donrCOt think thatrCOs been a big success ...
That's because God is my copilot ;-)
That's better than Copilot being your god, I guess.
On 11/13/26 00:31, Thomas Koenig wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-12, Alexander Schreiber <als@usenet.thangorodrim.de> wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous failures.
That's one of the things I like about aviation: it's one of
the few remaining places where safety is taken seriously.
I work in the chmemical industry, and we certainly take safety
seriously there.
But if you're interested in airline safety, I assume you read the
Aviation Herald? One disturbing trend is the increasing number
of lithium ion battery fires from things that passengers brought
on board.
I can't understand (well, I can...) why LiO2 batteries were ever
approved for general use. Just the other day I saw on the news that a
house burned (actually exploded) because if a battery fire.
On 11/13/26 00:31, Thomas Koenig wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-12, Alexander Schreiber <als@usenet.thangorodrim.de> wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous
failures.
That's one of the things I like about aviation: it's one of
the few remaining places where safety is taken seriously.
I work in the chmemical industry, and we certainly take safety
seriously there.
But if you're interested in airline safety, I assume you read the
Aviation Herald?-a One disturbing trend is the increasing number
of lithium ion battery fires from things that passengers brought
on board.
I can't understand (well, I can...) why LiO2 batteries were ever
approved for general use.
Just the other day I saw on the news that a
house burned (actually exploded) because if a battery fire.
An early Sinclair machine (ZX80? ZX81?) cut costs by having the screen refresh managed by the CPU in place of dedicated hardware. Every time
you pressed a key on the keyboard, the screen would go blank for a
fraction of a second, while the CPU stopped screen refresh to process
your input.
On 2026-11-12, c186282 <c186282@nnada.net> wrote:
On 11/11/26 22:44, Charlie Gibbs wrote:
On 2026-11-11, John Levine <johnl@taugh.com> wrote:
The two engine 777 had more advanced and reliable engines. Once the 777 >>>> was ceritfied for long trips over water (known as ETOPS) the 747 was
dead. The 777 carried nearly as many people but with two engines and
a two person flight crew.
ETOPS officially stood for "Extended Twin-engine OPerationS",
but it was colloquially known as "Engines Turn Or Passengers Swim".
Heh heh !
Frankly I'd be MUCH happier with at least THREE
engines if going over wide expanses.
That's why ETOPS was devised. It contains measures to improve
reliability. For instance, a different crew will work on each
engine, to prevent one group's possible systematic errors from
affecting the other.
Still, shit happens. The DC-10 in the Sioux City crash had
three separate hydraulic systems - and the disintegrating
engine took out all three. BTW that plane did have three
engines.
Lawrence DrCOOliveiro <ldo@nz.invalid> writes:
An early Sinclair machine (ZX80? ZX81?) cut costs by having the screen
refresh managed by the CPU in place of dedicated hardware. Every time
you pressed a key on the keyboard, the screen would go blank for a
fraction of a second, while the CPU stopped screen refresh to process
your input.
That's the ZX80, which uses an clever hack to generate a character
display with very minimal hardware using the Z80's instruction counter
and refresh counter to index the display RAM and character ROM -- so it
has to be running precisely-timed code (mostly a series of synthetic
NOPs) while it's generating video, and doing any other computation will
blank the screen, unless you're very careful to do it only during the vertical blanking.
The ZX81 works the same way, but has extra hardware to generate NMIs at
the line rate, so it can automatically switch to the display code when
the blanking is over. That's "slow mode", used by default, but you can
also switch it into ZX80-style "fast mode" if you want more CPU time and don't care about the display.
There's a good description here: https://blog.tynemouthsoftware.co.uk/2023/10/how-the-zx80-generates-video.html
Wholeheartedly agree. I'm very much a child of the commandline, my primary email client is mutt, my newsreader is slrn and my idea of a graphical user interface is "20+ xterms
and a few browser windows".
I refuse to touch a keyboard with an abomination called a "Windows
Key" on it.
The Windows key can be made useful by remapping it to Compose.
I've never found a useful way to use a "Window Key" with a Microsoft computer. With a Linux computer, this key is very useful. It's used toI use it for window-manager shortcuts for the same reason.
manage my windows knowing that every configuration I make with this
key won't mess with the shortcut of my programs.
Lawrence DrCOOliveiro <ldo@nz.invalid> schrieb:
So really, you are saying that general-purpose CPUs are now offloading
work ... onto other general-purpose CPUs.
Not "now".
Brooks wrote that one of the major design failures of the /360
was that they failed to realize their channel processors were,
in fact, computers.
On 13/11/2026 14:40, Peter Flass wrote:
On 11/13/26 00:31, Thomas Koenig wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-12, Alexander Schreiber <als@usenet.thangorodrim.de> wrote: >>>>
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous
failures.
That's one of the things I like about aviation: it's one of
the few remaining places where safety is taken seriously.
I work in the chmemical industry, and we certainly take safety
seriously there.
But if you're interested in airline safety, I assume you read the
Aviation Herald?-a One disturbing trend is the increasing number
of lithium ion battery fires from things that passengers brought
on board.
I can't understand (well, I can...) why LiO2 batteries were ever
approved for general use.
There is no such battery. Lithium Ion or LIPO for lithium ion polymer or LiFePo for Lithium iron polymer
Just the other day I saw on the news that a
house burned (actually exploded) because if a battery fire.
Batteries made badly and sold out of spec.
On 2026-11-13, Peter Flass <Peter@Iron-Spring.com> wrote:
On 11/13/26 00:31, Thomas Koenig wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-12, Alexander Schreiber <als@usenet.thangorodrim.de> wrote: >>>>
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous failures.
That's one of the things I like about aviation: it's one of
the few remaining places where safety is taken seriously.
I work in the chmemical industry, and we certainly take safety
seriously there.
But if you're interested in airline safety, I assume you read the
Aviation Herald? One disturbing trend is the increasing number
of lithium ion battery fires from things that passengers brought
on board.
I can't understand (well, I can...) why LiO2 batteries were ever
approved for general use. Just the other day I saw on the news that a
house burned (actually exploded) because if a battery fire.
Years and years ago, Transport Canada mandated that all aircraft
emergency locator transmitters (ELTs) must be equipped with lithium
batteries because they were the only ones guaranteed to work in
temperatures as low as -40 (Fahrenheit or Celsius, take your pick).
(Never mind the fact that few aircraft - especially the puddle-jumpers
in the general aviation fleet) were never taken that far north.)
So we all dutifully complied - and then the reports started coming
in of ELT batteries exploding, catching fire, or emitting harmful
gases.
So another directive came out mandating that all lithium
batteries be replaced (with magnesium batteries IIRC). When I
I pulled my ELT, I found corrosion around the connectors, from
gases that had managed to leak from inside the case.
On 11/10/26 10:14, Scott Lurndal wrote:
Pancho <Pancho.Jones@protonmail.com> writes:
On 11/10/26 09:37, Alexander Schreiber wrote:
No. The operating system and application software ran on the mainframe >>>> CPU. But - as described above - a key design aspect of mainframes back >>>> then was that I/O was not done by the CPU (like it was done on home"Fully offloaded" is a bit weasel worded. I thought PCs did offload some >>> duties to io devices, nic chip, wifi chip, gpu card. sound card. Also,
computers were all I/O was entirely handled by the CPU), but fully off- >>>> loaded to peripherials that were essentially limited computers of their >>>
even a mainframe will need service routines to handle communications
with IO devices. Where to draw the line seems unclear.
Fully offloaded is accurate.-a For the Burroughs mainframes, a single
instruction -initiated- an I/O request to a channel.-a-a The IIO (later
SPIO) instruction would instruct the I/O processor (in the late 70s
the IOP used an 8085 uProcessor) to initiate an I/O by passing a
6-digit (24-bit) opcode/variant word and the start and end addresses
of the buffer.-a-a The IOP would pass the command to a Data-Link Processor >> (DLP) that handled communications to the device controller (magnetic
tape, disk) or unit record device (printer, card reader, card punch)
or external communications processor (RS232/20ma current loop) and wait
for the controller to start passing data back to the IOP.-a The IOP
would move the data to the appropriate memory buffer and, when complete,
signal the processor that the I/O had completed with an interrupt.
-a Esp in the Bad Old Days, "smart perips" were VERY
-a important. The CPU just sent a LITTLE stuff to them.
-a Very CPU/RAM/Cycle efficient.
-a Flip - left you hostage to the periph "IQ", often
-a no ways to tweak/enhance it all.
Caveat: I don't have a clue about these things, it is just that what you >>> are saying doesn't sound quite right.
-a Me ... SEEN it all - the plusses AND minuses.
-a The somewhat older group members will have
-a seen MORE of that.
-a The newbies think NOTHING from the past can
-a possibly inform them, pave the way.
-a Horribly wrong ... but you'll never convince them.
-a For them it's now all "AI" ... until it displaces
-a them entirely.
-a Can ya say "Would You Like Fries With That ?" in
-a three or four languages ???-a :-)
... lithium battery chemistries instead of with much safer ones e.g.
LiFePo4.
On Fri, 13 Nov 2026 22:02:26 +0100, Alexander Schreiber wrote:
... lithium battery chemistries instead of with much safer ones e.g.
LiFePo4.
Polonium?!?
The Natural Philosopher <tnp@invalid.invalid> wrote:
On 13/11/2026 14:40, Peter Flass wrote:
On 11/13/26 00:31, Thomas Koenig wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-12, Alexander Schreiber <als@usenet.thangorodrim.de> wrote: >>>>>
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous >>>>>> failures.
That's one of the things I like about aviation: it's one of
the few remaining places where safety is taken seriously.
I work in the chmemical industry, and we certainly take safety
seriously there.
But if you're interested in airline safety, I assume you read the
Aviation Herald?-a One disturbing trend is the increasing number
of lithium ion battery fires from things that passengers brought
on board.
I can't understand (well, I can...) why LiO2 batteries were ever
approved for general use.
There is no such battery. Lithium Ion or LIPO for lithium ion polymer or
LiFePo for Lithium iron polymer
Well, there are Lithium-Air batteries. But while various implementations
of them have been demonstrated, AFAIK they are nowhere near commercialization.
Just the other day I saw on the news that a
house burned (actually exploded) because if a battery fire.
Batteries made badly and sold out of spec.
And/or badly abused during transport, installation, operation. Plus
people with little knowledge of the technology tend to end up with
the more exciting (read: fire-prone) lithium battery chemistries instead
of with much safer ones e.g. LiFePo4. It doesn't help that often the
cell chemistry is not explicitly mentioned, e.g. with battery packs
for e-bikes and such.
Kind regards,
Alex.
On 11/11/2026 12:46 PM, c186282 wrote:
On 11/10/26 10:14, Scott Lurndal wrote:
Pancho <Pancho.Jones@protonmail.com> writes:
On 11/10/26 09:37, Alexander Schreiber wrote:
No. The operating system and application software ran on the mainframe >>>>> CPU. But - as described above - a key design aspect of mainframes back >>>>> then was that I/O was not done by the CPU (like it was done on home
computers were all I/O was entirely handled by the CPU), but fully
off-
loaded to peripherials that were essentially limited computers of
their
"Fully offloaded" is a bit weasel worded. I thought PCs did offload
some
duties to io devices, nic chip, wifi chip, gpu card. sound card. Also, >>>> even a mainframe will need service routines to handle communications
with IO devices. Where to draw the line seems unclear.
Fully offloaded is accurate.-a For the Burroughs mainframes, a single
instruction -initiated- an I/O request to a channel.-a-a The IIO (later
SPIO) instruction would instruct the I/O processor (in the late 70s
the IOP used an 8085 uProcessor) to initiate an I/O by passing a
6-digit (24-bit) opcode/variant word and the start and end addresses
of the buffer.-a-a The IOP would pass the command to a Data-Link Processor >>> (DLP) that handled communications to the device controller (magnetic
tape, disk) or unit record device (printer, card reader, card punch)
or external communications processor (RS232/20ma current loop) and wait
for the controller to start passing data back to the IOP.-a The IOP
would move the data to the appropriate memory buffer and, when complete, >>> signal the processor that the I/O had completed with an interrupt.
-a-a Esp in the Bad Old Days, "smart perips" were VERY
-a-a important. The CPU just sent a LITTLE stuff to them.
-a-a Very CPU/RAM/Cycle efficient.
-a-a Flip - left you hostage to the periph "IQ", often
-a-a no ways to tweak/enhance it all.
Caveat: I don't have a clue about these things, it is just that what
you
are saying doesn't sound quite right.
-a-a Me ... SEEN it all - the plusses AND minuses.
-a-a The somewhat older group members will have
-a-a seen MORE of that.
-a-a The newbies think NOTHING from the past can
-a-a possibly inform them, pave the way.
-a-a Horribly wrong ... but you'll never convince them.
-a-a For them it's now all "AI" ... until it displaces
-a-a them entirely.
1970s "cloud computing" would be lugging around a VT-of-the-day, and a modem.-a Then you could connect it to the hotel's phone line and get some work done while far and away.
Or at least, that's what I imagine it could have been; I lack the per-
sonal experience of it.
But no, no; that's all /new/.-a Cloud computing is hip and cool.
-a-a Can ya say "Would You Like Fries With That ?" in
-a-a three or four languages ???-a :-)
Yes, as a matter of fact I can; English, Icelandic, and French.-a Proba-
bly more languages in the future.
Best wishes, and happy terminals!
%! P.S. mine is VT520, so not a 70s vintage.
On Fri, 13 Nov 2026 22:02:26 +0100, Alexander Schreiber wrote:
... lithium battery chemistries instead of with much safer ones e.g.
LiFePo4.
Polonium?!?
Le 12-11-2026, Alexander Schreiber <als@usenet.thangorodrim.de> a |-crit-a:
The Windows key can be made useful by remapping it to Compose.
Yes, but I don't consider it a remap.
On 13 Nov 2026 19:58:39 GMT
St|-phane CARPENTIER <sc@fiat-linux.fr> wrote:
I've never found a useful way to use a "Window Key" with a Microsoft
computer. With a Linux computer, this key is very useful. It's used to
manage my windows knowing that every configuration I make with this
key won't mess with the shortcut of my programs.
I use it for window-manager shortcuts for the same reason.
The Natural Philosopher <tnp@invalid.invalid> wrote:
On 13/11/2026 14:40, Peter Flass wrote:
On 11/13/26 00:31, Thomas Koenig wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-12, Alexander Schreiber <als@usenet.thangorodrim.de> wrote: >>>>>
Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
Many redundant aircraft systems are routed (or re-routed if
a potential problem is later discovered) to try to prevent
one incident from damaging them all.
Indeed. At least aviation is pretty good at learning from previous >>>>>> failures.
That's one of the things I like about aviation: it's one of
the few remaining places where safety is taken seriously.
I work in the chmemical industry, and we certainly take safety
seriously there.
But if you're interested in airline safety, I assume you read the
Aviation Herald?-a One disturbing trend is the increasing number
of lithium ion battery fires from things that passengers brought
on board.
I can't understand (well, I can...) why LiO2 batteries were ever
approved for general use.
There is no such battery. Lithium Ion or LIPO for lithium ion polymer or
LiFePo for Lithium iron polymer
Well, there are Lithium-Air batteries. But while various implementations
of them have been demonstrated, AFAIK they are nowhere near commercialization.
All of the above.Just the other day I saw on the news that a
house burned (actually exploded) because if a battery fire.
Batteries made badly and sold out of spec.
And/or badly abused during transport, installation, operation. Plus
people with little knowledge of the technology tend to end up with
the more exciting (read: fire-prone) lithium battery chemistries instead
of with much safer ones e.g. LiFePo4. It doesn't help that often the
cell chemistry is not explicitly mentioned, e.g. with battery packs
for e-bikes and such.
Kind regards,
Alex.
On 11/13/26 20:09, Lawrence DrCOOliveiro wrote:
On Fri, 13 Nov 2026 22:02:26 +0100, Alexander Schreiber wrote:
... lithium battery chemistries instead of with much safer ones e.g.
LiFePo4.
Polonium?!?
-a :-)
-a Sometimes capitalization does count.
-a DO remember when they promised everything would
-a be run by atomic batteries that'd last 100 years.
Lars Poulsen <lars@beagle-ears.com> wrote:
On a transoceanic route, time in the air is quite significant.
On a short "puddlejumper" route, a high-speed train (TGV in French) may
well be faster.
In fact the rollout of the TGV network largely killed off domestic flights
in France, e.g. the Paris-Lyon TGV line opening in 1981 caused domestic
air travel on that corridor to drop by 60% in one year.
And once you factor
in all the overhead (not flying) time of air travel, taking a train can actually be faster on shorter (domestic) routes. It tends to be more convenient too.
1970s "cloud computing" would be lugging around a VT-of-the-day, and a modem.-a Then you could connect it to the hotel's phone line and get some work done while far and away.
Or at least, that's what I imagine it could have been; I lack the per-
sonal experience of it.
But no, no; that's all /new/.-a Cloud computing is hip and cool.
%! P.S. mine is VT520, so not a 70s vintage.
So, the plane is faster than the TGV, but it takes more time to go from
Paris to Lyon on plane than on TGV. Maybe that's why I never heard about anyone taking the plane to go from Paris to Lyon. It's neither cheaper
nor more comfortable. Maybe it can be taken by someone coming from far
away, with no plane to Lyon and whose only possibility is to go to
Paris. It can be an opportunity to take the plane directly from the
airport but I'm not entirely convinced.
On 11/13/26 17:24, Johann 'Myrkraverk' Oskarsson wrote:
On 11/11/2026 12:46 PM, c186282 wrote:
On 11/10/26 10:14, Scott Lurndal wrote:
Pancho <Pancho.Jones@protonmail.com> writes:
On 11/10/26 09:37, Alexander Schreiber wrote:
No. The operating system and application software ran on the
mainframe
CPU. But - as described above - a key design aspect of mainframes >>>>>> back
then was that I/O was not done by the CPU (like it was done on home >>>>>> computers were all I/O was entirely handled by the CPU), but fully >>>>>> off-
loaded to peripherials that were essentially limited computers of >>>>>> their
"Fully offloaded" is a bit weasel worded. I thought PCs did offload >>>>> some
duties to io devices, nic chip, wifi chip, gpu card. sound card. Also, >>>>> even a mainframe will need service routines to handle communications >>>>> with IO devices. Where to draw the line seems unclear.
Fully offloaded is accurate.-a For the Burroughs mainframes, a single
instruction -initiated- an I/O request to a channel.-a-a The IIO (later >>>> SPIO) instruction would instruct the I/O processor (in the late 70s
the IOP used an 8085 uProcessor) to initiate an I/O by passing a
6-digit (24-bit) opcode/variant word and the start and end addresses
of the buffer.-a-a The IOP would pass the command to a Data-Link
Processor
(DLP) that handled communications to the device controller (magnetic
tape, disk) or unit record device (printer, card reader, card punch)
or external communications processor (RS232/20ma current loop) and wait >>>> for the controller to start passing data back to the IOP.-a The IOP
would move the data to the appropriate memory buffer and, when
complete,
signal the processor that the I/O had completed with an interrupt.
-a-a Esp in the Bad Old Days, "smart perips" were VERY
-a-a important. The CPU just sent a LITTLE stuff to them.
-a-a Very CPU/RAM/Cycle efficient.
-a-a Flip - left you hostage to the periph "IQ", often
-a-a no ways to tweak/enhance it all.
Caveat: I don't have a clue about these things, it is just that
what you
are saying doesn't sound quite right.
-a-a Me ... SEEN it all - the plusses AND minuses.
-a-a The somewhat older group members will have
-a-a seen MORE of that.
-a-a The newbies think NOTHING from the past can
-a-a possibly inform them, pave the way.
-a-a Horribly wrong ... but you'll never convince them.
-a-a For them it's now all "AI" ... until it displaces
-a-a them entirely.
1970s "cloud computing" would be lugging around a VT-of-the-day, and a
modem.-a Then you could connect it to the hotel's phone line and get some
work done while far and away.
Or at least, that's what I imagine it could have been; I lack the per-
sonal experience of it.
But no, no; that's all /new/.-a Cloud computing is hip and cool.
-a-a Can ya say "Would You Like Fries With That ?" in
-a-a three or four languages ???-a :-)
Yes, as a matter of fact I can; English, Icelandic, and French.-a Proba-
bly more languages in the future.
Best wishes, and happy terminals!
%! P.S. mine is VT520, so not a 70s vintage.
-a After all the prose, it has become clear that
-a "mainframe" is very hard to define. There were
-a many players, many models, everything a bit
-a different, different emphasis on various
-a aspects.
FAX was the closest
thing to email.
In contrast, I distinctly remember that we had dial-in lines on the UNIX systems at ACC when I got to California. After a while (1981?), I got an LS-120 at home; at first I used an acoustic coupler, but soon after
that, I got a modem; probably a 300 bps.
Le 12-11-2026, Alexander Schreiber <als@usenet.thangorodrim.de> a |-crit-a:
Lars Poulsen <lars@beagle-ears.com> wrote:
On a transoceanic route, time in the air is quite significant.
On a short "puddlejumper" route, a high-speed train (TGV in French) may
well be faster.
In fact the rollout of the TGV network largely killed off domestic flights >> in France, e.g. the Paris-Lyon TGV line opening in 1981 caused domestic
air travel on that corridor to drop by 60% in one year.
I never heard about anyone taking a plane from Paris to Lyon. I didn't
knew that was possible, so I just checked and it still exists.
And once you factor
in all the overhead (not flying) time of air travel, taking a train can
actually be faster on shorter (domestic) routes. It tends to be more
convenient too.
Now a TVG takes around 2h00 to go from Paris to Lyon. And a plane takes something like 1h10. But, as you said, you have to take the onboarding
time which is far longer on a plane than on a train. And the Charles de Gaulle airport is not in Paris but is at something like 30km from it.
And the Saint Exup|-ry airport is not in Lyon but is at something like
30km from it.
So, the plane is faster than the TGV, but it takes more time to go from
Paris to Lyon on plane than on TGV. Maybe that's why I never heard about anyone taking the plane to go from Paris to Lyon. It's neither cheaper
nor more comfortable. Maybe it can be taken by someone coming from far
away, with no plane to Lyon and whose only possibility is to go to
Paris. It can be an opportunity to take the plane directly from the
airport but I'm not entirely convinced.
I read somewhere that /mainframe/ got its name from the -- wait for it
-- main frame of the large vacuum tube computers of the 1950s, if not
40s.-a I don't recall seeing pictures of these /main frames/, so I don't
know what they looked like.
It's just one of these computing terms that's not quite shrouded in mys---
tery but few people know the etymology of it.
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it
-- main frame of the large vacuum tube computers of the 1950s, if not
40s.-a I don't recall seeing pictures of these /main frames/, so I don't
know what they looked like.
a computer built in multiple 19" racks?
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it
-- main frame of the large vacuum tube computers of the 1950s, if not
40s.-a I don't recall seeing pictures of these /main frames/, so I don't >>> know what they looked like.
a computer built in multiple 19" racks?
What about all those DEC minis built in 19" racks?
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames. Much more exclusive and sexy.
In fact the rollout of the TGV network largely killed off domestic flights >>> in France, e.g. the Paris-Lyon TGV line opening in 1981 caused domestic
air travel on that corridor to drop by 60% in one year.
I never heard about anyone taking a plane from Paris to Lyon. I didn't
knew that was possible, so I just checked and it still exists.
Based on personal experience, these flights serve the transfer passen-
gers mostly [1]. I doubt people traveling from Paris to Lyon would
choose flight over the TGV /just because/. But New York to Lyon is a >different story.
After a while (1981?), I got an LS-120 at home ...
It appears that Charlie Gibbs <cgibbs@kltpzyxm.invalid> said:
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it >>>> -- main frame of the large vacuum tube computers of the 1950s, if not
40s.-a I don't recall seeing pictures of these /main frames/, so I don't >>>> know what they looked like.
a computer built in multiple 19" racks?
Yes indeed, adapted from racks used in phone exchanges.
What about all those DEC minis built in 19" racks?
Well, sure, if it was too big to sit on a desk they might as well
use (so to speak) off the shelf racks.
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames. Much more exclusive and sexy.
I think that is also to minimize cable lengths, and probably has
something to do with the piping for water cooling.
On 11/14/2026 1:33 PM, c186282 wrote:
On 11/13/26 17:24, Johann 'Myrkraverk' Oskarsson wrote:
On 11/11/2026 12:46 PM, c186282 wrote:
On 11/10/26 10:14, Scott Lurndal wrote:
Pancho <Pancho.Jones@protonmail.com> writes:
On 11/10/26 09:37, Alexander Schreiber wrote:
No. The operating system and application software ran on the
mainframe
CPU. But - as described above - a key design aspect of mainframes >>>>>>> back
then was that I/O was not done by the CPU (like it was done on home >>>>>>> computers were all I/O was entirely handled by the CPU), but
fully off-
loaded to peripherials that were essentially limited computers of >>>>>>> their
"Fully offloaded" is a bit weasel worded. I thought PCs did
offload some
duties to io devices, nic chip, wifi chip, gpu card. sound card.
Also,
even a mainframe will need service routines to handle communications >>>>>> with IO devices. Where to draw the line seems unclear.
Fully offloaded is accurate.-a For the Burroughs mainframes, a single >>>>> instruction -initiated- an I/O request to a channel.-a-a The IIO (later >>>>> SPIO) instruction would instruct the I/O processor (in the late 70s
the IOP used an 8085 uProcessor) to initiate an I/O by passing a
6-digit (24-bit) opcode/variant word and the start and end addresses >>>>> of the buffer.-a-a The IOP would pass the command to a Data-Link
Processor
(DLP) that handled communications to the device controller (magnetic >>>>> tape, disk) or unit record device (printer, card reader, card punch) >>>>> or external communications processor (RS232/20ma current loop) and
wait
for the controller to start passing data back to the IOP.-a The IOP
would move the data to the appropriate memory buffer and, when
complete,
signal the processor that the I/O had completed with an interrupt.
-a-a Esp in the Bad Old Days, "smart perips" were VERY
-a-a important. The CPU just sent a LITTLE stuff to them.
-a-a Very CPU/RAM/Cycle efficient.
-a-a Flip - left you hostage to the periph "IQ", often
-a-a no ways to tweak/enhance it all.
Caveat: I don't have a clue about these things, it is just that
what you
are saying doesn't sound quite right.
-a-a Me ... SEEN it all - the plusses AND minuses.
-a-a The somewhat older group members will have
-a-a seen MORE of that.
-a-a The newbies think NOTHING from the past can
-a-a possibly inform them, pave the way.
-a-a Horribly wrong ... but you'll never convince them.
-a-a For them it's now all "AI" ... until it displaces
-a-a them entirely.
1970s "cloud computing" would be lugging around a VT-of-the-day, and a
modem.-a Then you could connect it to the hotel's phone line and get some >>> work done while far and away.
Or at least, that's what I imagine it could have been; I lack the per-
sonal experience of it.
But no, no; that's all /new/.-a Cloud computing is hip and cool.
-a-a Can ya say "Would You Like Fries With That ?" in
-a-a three or four languages ???-a :-)
Yes, as a matter of fact I can; English, Icelandic, and French.-a Proba- >>> bly more languages in the future.
Best wishes, and happy terminals!
%! P.S. mine is VT520, so not a 70s vintage.
-a-a After all the prose, it has become clear that
-a-a "mainframe" is very hard to define. There were
-a-a many players, many models, everything a bit
-a-a different, different emphasis on various
-a-a aspects.
I read somewhere that /mainframe/ got its name from the -- wait for it
-- main frame of the large vacuum tube computers of the 1950s, if not
40s.-a I don't recall seeing pictures of these /main frames/, so I don't
know what they looked like.
It's just one of these computing terms that's not quite shrouded in mys-
tery but few people know the etymology of it.
Best wishes, and happy mainframes!
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
Perception doesnrCOt count. ThatrCOs reality. You canrCOt *pretend* to
offer interactivity: you really have to have support for
interactivity built into your architecture from the lowest levels
up. You canrCOt build per-character response on top of block mode.
You absolutely can and the interactive web proves it.
WebSocket connections are asynchronous full-duplex connections --
thatrCOs not rCLRPCrCY by any stretch.
Of course it is. RPCs come in various kinds, from the classic
one-shot (establish connection, send request, wait for and read
response, close connection) to streaming RPCs.
Even HTTP, which is not exactly the most efficient RPC method, came
up with streaming RPCs: HTTP 1.1 pipelining, where multiple requests
are handled within the same TCP connection.
And "asynchronous" just means you don't do blocking reads and
instead use the framework equivalent of select(2) for a more fluid interaction.
And nobody ever described block-mode mainframe terminals as running any
kind of rCLRPCrCY either. The rCLRPCrCY concept was only invented after mainframes had been around for a few decades.
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it
-- main frame of the large vacuum tube computers of the 1950s, if not
40s.-a I don't recall seeing pictures of these /main frames/, so I don't >>> know what they looked like.
a computer built in multiple 19" racks?
What about all those DEC minis built in 19" racks?
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames. Much more exclusive and sexy.
Whee, this is becoming fun...
I never worked on water-cooled machines, but the swing-out frames
allowed a lot of circuitry to be packed into a box, and gave easy
access to everything.
On Sun, 15 Nov 2026 05:57:04 -0000 (UTC), Lawrence DrCOOliveiro wrote:
And nobody ever described block-mode mainframe terminals as running any
kind of rCLRPCrCY either. The rCLRPCrCY concept was only invented after
mainframes had been around for a few decades.
I don't have a timeline but I mostly worked with ONC-RPC, which was a Sun protocol from the '80s. XDR had its oddities but was fairly straight
forward. In later years using network byte order on x86 only systems was slightly inefficient.
Next was XML-RPC that grew into SOAP and was supported by the wondrous Windows Communication Framework. imnsho it was a steaming pile of shit and
if I never see another wsdl I'll be happy.
Then we moved on to an Angular SPA and I learned the joys of protobufs.
From my viewpoint that was almost back to ONC-RPC without all the
portmapper complexity. We never used the word gRPC but it was probably
close.
Then there is the who DCE/RPC family that Microsoft derived MSRPC from and used for COM and DCOM.
Many strange flavors of RPC.
On 14/11/2026 18:34, Charlie Gibbs wrote:
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:Indeed. Originally it was what it was, but it became
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it >>>> -- main frame of the large vacuum tube computers of the 1950s, if not
40s.-a I don't recall seeing pictures of these /main frames/, so I don't >>>> know what they looked like.
a computer built in multiple 19" racks?
What about all those DEC minis built in 19" racks?
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames.-a Much more exclusive and sexy.
Whee, this is becoming fun...
1. A computer able to handled thousands of user sessions
or
2.-a A computer with massive storage and the ability to do payroll
batches in time for payday.
Naturally these are very large organisation solutions, so not many
people were exposed to them. Which is why so many people think these machines no longer exist
JSON is probably the most popular today.-a I personally don't like it for being wasteful.
On 11/15/2026 5:52 PM, The Natural Philosopher wrote:
On 14/11/2026 18:34, Charlie Gibbs wrote:
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:Indeed. Originally it was what it was, but it became
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it >>>>> -- main frame of the large vacuum tube computers of the 1950s, if not >>>>> 40s.-a I don't recall seeing pictures of these /main frames/, so I
don't
know what they looked like.
a computer built in multiple 19" racks?
What about all those DEC minis built in 19" racks?
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames.-a Much more exclusive and sexy.
Whee, this is becoming fun...
1. A computer able to handled thousands of user sessions
or
2.-a A computer with massive storage and the ability to do payroll
batches in time for payday.
Naturally these are very large organisation solutions, so not many
people were exposed to them. Which is why so many people think these
machines no longer exist
The problem of /how do we get $amount into every employee's bank ac-
count/ with thousands of employees and perhaps a single payroll account
for the company does not sound like a solution waiting for a cluster of
Linux boxes.
On 11/15/2026 5:52 PM, The Natural Philosopher wrote:
On 14/11/2026 18:34, Charlie Gibbs wrote:
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:Indeed. Originally it was what it was, but it became
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it >>>>> -- main frame of the large vacuum tube computers of the 1950s, if not >>>>> 40s.-a I don't recall seeing pictures of these /main frames/, so I don't >>>>> know what they looked like.
a computer built in multiple 19" racks?
What about all those DEC minis built in 19" racks?
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames.-a Much more exclusive and sexy.
Whee, this is becoming fun...
1. A computer able to handled thousands of user sessions
or
2.-a A computer with massive storage and the ability to do payroll
batches in time for payday.
Naturally these are very large organisation solutions, so not many
people were exposed to them. Which is why so many people think these
machines no longer exist
The problem of /how do we get $amount into every employee's bank ac-
count/ with thousands of employees and perhaps a single payroll account
for the company does not sound like a solution waiting for a cluster of
Linux boxes.
On 11/13/26 16:02, Alexander Schreiber wrote:[]
The Natural Philosopher <tnp@invalid.invalid> wrote:
On 13/11/2026 14:40, Peter Flass wrote:
I can't understand (well, I can...) why LiO2 batteries were ever
approved for general use.
There is no such battery. Lithium Ion or LIPO for lithium ion polymer or >> LiFePo for Lithium iron polymer
Well, there are Lithium-Air batteries. But while various implementations
of them have been demonstrated, AFAIK they are nowhere near commercialization.
Just the other day I saw on the news that a
house burned (actually exploded) because if a battery fire.
Batteries made badly and sold out of spec.
And/or badly abused during transport, installation, operation. Plus
people with little knowledge of the technology tend to end up with
the more exciting (read: fire-prone) lithium battery chemistries instead
of with much safer ones e.g. LiFePo4. It doesn't help that often the
cell chemistry is not explicitly mentioned, e.g. with battery packs
for e-bikes and such.
The best alt these days are sodium batteries ... already
commercial, you can buy ones meant to start up automobiles
on Amazon. Not quite the power density, but getting close
and don't explode.
As for 'quality batteries', just no WAY to know what
you're really getting these days.
On 2026-11-15, Johann 'Myrkraverk' Oskarsson <johann@myrkraverk.invalid> wrote:
On 11/15/2026 5:52 PM, The Natural Philosopher wrote:
On 14/11/2026 18:34, Charlie Gibbs wrote:
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:Indeed. Originally it was what it was, but it became
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it >>>>>> -- main frame of the large vacuum tube computers of the 1950s, if not >>>>>> 40s.-a I don't recall seeing pictures of these /main frames/, so I don't >>>>>> know what they looked like.
a computer built in multiple 19" racks?
What about all those DEC minis built in 19" racks?
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames.-a Much more exclusive and sexy.
Whee, this is becoming fun...
1. A computer able to handled thousands of user sessions
or
2.-a A computer with massive storage and the ability to do payroll
batches in time for payday.
Naturally these are very large organisation solutions, so not many
people were exposed to them. Which is why so many people think these
machines no longer exist
The problem of /how do we get $amount into every employee's bank ac-
count/ with thousands of employees and perhaps a single payroll account
for the company does not sound like a solution waiting for a cluster of
Linux boxes.
The problem is actually how do we get the *right* amount into every employee's bank account (or charge the *right* amount of tax, or give
the *right* amount of benefits...), where *right* is defined by years
or decades of ad-hoc business logic and legislation that is embedded into ancient software (COBOL/JCL et. al.) that can only run on the "mainframe"
it was developed on.
Too often the shiny new solution doesn't get the right answers, despite
being infinitely faster. So what do you do - keep paying your mainframe supplier and not have any risk, or switch to a modern solution and risk
your job if thousands of people complain that you've got the sums wrong?
Yes, eventually the modern solution wins, but IMO this is the primary
reason "mainframes" still exist.
So today, "mainframe" = "System capable of running software developed
before Linux or Windows was ever conceived" :)
On Fri, 13 Nov 2026 15:05:00 +0100, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
Perception doesnrCOt count. ThatrCOs reality. You canrCOt *pretend* to
offer interactivity: you really have to have support for
interactivity built into your architecture from the lowest levels
up. You canrCOt build per-character response on top of block mode.
You absolutely can and the interactive web proves it.
The interactive web is not run on mainframe-style block-mode
terminals.
WebSocket connections are asynchronous full-duplex connections --
thatrCOs not rCLRPCrCY by any stretch.
Of course it is. RPCs come in various kinds, from the classic
one-shot (establish connection, send request, wait for and read
response, close connection) to streaming RPCs.
rCLRPCrCY stands for rCLRemote Procedure CallrCY. ItrCOs emulating a procedure
call over a network connection. You know how a procedure call works,
donrCOt you? The caller transfers control to the callee, waits for the
callee to do its thing and return a result, whereupon the caller
resumes execution again.
And nobody ever described block-mode mainframe terminals as running
any kind of rCLRPCrCY either. The rCLRPCrCY concept was only invented after mainframes had been around for a few decades.
Before getting even deeper into your love for RPCs and seeing every
kind of network protocol as RPCs, maybe you should read this
<https://www.cs.vu.nl/~ast/Publications/Papers/euteco-1988.pdf>.
Even HTTP, which is not exactly the most efficient RPC method, came
up with streaming RPCs: HTTP 1.1 pipelining, where multiple requests
are handled within the same TCP connection.
Those donrCOt work for RPCs, though.
And "asynchronous" just means you don't do blocking reads and
instead use the framework equivalent of select(2) for a more fluid
interaction.
Well, maybe not select(2), maybe something more modern?
Yes, eventually the modern solution wins, but IMO this is the primary
reason "mainframes" still exist.
So today, "mainframe" = "System capable of running software developed
before Linux or Windows was ever conceived" EfOe
Bottom line, do we care if we run the Linux on a mainframe?
In retrospect, there are
functional similarities between those block mode terminals and web
services, which might be more obvious if one uses a terminal browser
like links for the web (which of course limits one to sites that don't
rely on JavaScript and lots of graphics for their function).
The problem of /how do we get $amount into every employee's bank ac-
count/ with thousands of employees and perhaps a single payroll account
for the company does not sound like a solution waiting for a cluster of
Linux boxes.
On 15/11/2026 12:22, Ian wrote:
Yes, eventually the modern solution wins, but IMO this is the primary
reason "mainframes" still exist.
So today, "mainframe" = "System capable of running software developed
before Linux or Windows was ever conceived" :)
Simply not true
All trad OS and languages run on linux machines of the standard sort - Fortran - COBOL - RPG...SQL.
The problem is actually how do we get the *right* amount into every employee's bank account (or charge the *right* amount of tax, or
give the *right* amount of benefits...), where *right* is defined by
years or decades of ad-hoc business logic and legislation that is
embedded into ancient software (COBOL/JCL et. al.) that can only run
on the "mainframe" it was developed on.
Too often the shiny new solution doesn't get the right answers,
despite being infinitely faster. So what do you do - keep paying
your mainframe supplier and not have any risk, or switch to a modern
solution and risk your job if thousands of people complain that
you've got the sums wrong?
Yes, eventually the modern solution wins, but IMO this is the primary
reason "mainframes" still exist.
So today, "mainframe" = "System capable of running software developed
before Linux or Windows was ever conceived" :)
Johann 'Myrkraverk' Oskarsson <johann@myrkraverk.invalid> schrieb:
The problem of /how do we get $amount into every employee's bank ac-
count/ with thousands of employees and perhaps a single payroll account
for the company does not sound like a solution waiting for a cluster of
Linux boxes.
SAP moved away from mainframes in the early 1990s with SAP R3, and
they do run payrolls.
And SAP S4/HANA runs on... you guessed it, Linux, exclusively.
It does not even run on mainframes any more - it's Intel (not AMD)
or IBM POWER.
JSON is probably the most popular today. I personally don't like it for being wasteful.
On 15/11/2026 10:04, Johann 'Myrkraverk' Oskarsson wrote:
JSON is probably the most popular today.-a I personally don't like it
for being wasteful.
There are enough bugs in JavaScript without introducing yet another
layer...
I use the minimum pure JavaScript sufficient to get the job done
On 2026-11-15, Johann 'Myrkraverk' Oskarsson <johann@myrkraverk.invalid> wrote:
On 11/15/2026 5:52 PM, The Natural Philosopher wrote:
On 14/11/2026 18:34, Charlie Gibbs wrote:
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:Indeed. Originally it was what it was, but it became
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it >>>>>> -- main frame of the large vacuum tube computers of the 1950s, if not >>>>>> 40s.-a I don't recall seeing pictures of these /main frames/, so I don't >>>>>> know what they looked like.
a computer built in multiple 19" racks?
What about all those DEC minis built in 19" racks?
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames.-a Much more exclusive and sexy.
Whee, this is becoming fun...
1. A computer able to handled thousands of user sessions
or
2.-a A computer with massive storage and the ability to do payroll
batches in time for payday.
Naturally these are very large organisation solutions, so not many
people were exposed to them. Which is why so many people think these
machines no longer exist
The problem of /how do we get $amount into every employee's bank ac-
count/ with thousands of employees and perhaps a single payroll account
for the company does not sound like a solution waiting for a cluster of
Linux boxes.
The problem is actually how do we get the *right* amount into every employee's bank account (or charge the *right* amount of tax, or give
the *right* amount of benefits...), where *right* is defined by years
or decades of ad-hoc business logic and legislation that is embedded into ancient software (COBOL/JCL et. al.) that can only run on the "mainframe"
it was developed on.
Too often the shiny new solution doesn't get the right answers, despite
being infinitely faster. So what do you do - keep paying your mainframe supplier and not have any risk, or switch to a modern solution and risk
your job if thousands of people complain that you've got the sums wrong?
Yes, eventually the modern solution wins, but IMO this is the primary
reason "mainframes" still exist.
So today, "mainframe" = "System capable of running software developed
before Linux or Windows was ever conceived" :)
On 11/15/2026 5:52 PM, The Natural Philosopher wrote:
On 14/11/2026 18:34, Charlie Gibbs wrote:
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:Indeed. Originally it was what it was, but it became
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it >>>>> -- main frame of the large vacuum tube computers of the 1950s, if not >>>>> 40s.-a I don't recall seeing pictures of these /main frames/, so I don't >>>>> know what they looked like.
a computer built in multiple 19" racks?
What about all those DEC minis built in 19" racks?
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames.-a Much more exclusive and sexy.
Whee, this is becoming fun...
1. A computer able to handled thousands of user sessions
or
2.-a A computer with massive storage and the ability to do payroll
batches in time for payday.
Naturally these are very large organisation solutions, so not many
people were exposed to them. Which is why so many people think these
machines no longer exist
The problem of /how do we get $amount into every employee's bank ac-
count/ with thousands of employees and perhaps a single payroll account
for the company does not sound like a solution waiting for a cluster of
Linux boxes.
On 15/11/2026 10:33, Johann 'Myrkraverk' Oskarsson wrote:
On 11/15/2026 5:52 PM, The Natural Philosopher wrote:
On 14/11/2026 18:34, Charlie Gibbs wrote:
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:Indeed. Originally it was what it was, but it became
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it >>>>>> -- main frame of the large vacuum tube computers of the 1950s, if not >>>>>> 40s.-a I don't recall seeing pictures of these /main frames/, so I >>>>>> don't
know what they looked like.
a computer built in multiple 19" racks?
What about all those DEC minis built in 19" racks?
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames.-a Much more exclusive and sexy.
Whee, this is becoming fun...
1. A computer able to handled thousands of user sessions
or
2.-a A computer with massive storage and the ability to do payroll
batches in time for payday.
Naturally these are very large organisation solutions, so not many
people were exposed to them. Which is why so many people think these
machines no longer exist
The problem of /how do we get $amount into every employee's bank ac-
count/ with thousands of employees and perhaps a single payroll account
for the company does not sound like a solution waiting for a cluster of
Linux boxes.
That was essentially my point.
Neither is necessarily running a global airline booking application.
We have a range of tools at our disposal, from a Raspberry Pi Pico to an
IBM mainframe, and none of them are suitable for *all* applications.
In the middle, a linux *86 style machine is a very useful and cost
effective machine, but not at the spectrum edges.
The Natural Philosopher <tnp@invalid.invalid> wrote:
On 15/11/2026 10:33, Johann 'Myrkraverk' Oskarsson wrote:
On 11/15/2026 5:52 PM, The Natural Philosopher wrote:
On 14/11/2026 18:34, Charlie Gibbs wrote:
On 2026-11-14, The Natural Philosopher <tnp@invalid.invalid> wrote:Indeed. Originally it was what it was, but it became
On 14/11/2026 15:12, Johann 'Myrkraverk' Oskarsson wrote:
I read somewhere that /mainframe/ got its name from the -- wait for it >>>>>>> -- main frame of the large vacuum tube computers of the 1950s, if not >>>>>>> 40s.-a I don't recall seeing pictures of these /main frames/, so I >>>>>>> don't
know what they looked like.
a computer built in multiple 19" racks?
What about all those DEC minis built in 19" racks?
Many mainframes were built in bespoke racks, often with
backplanes on swing-out frames.-a Much more exclusive and sexy.
Whee, this is becoming fun...
1. A computer able to handled thousands of user sessions
or
2.-a A computer with massive storage and the ability to do payroll
batches in time for payday.
Naturally these are very large organisation solutions, so not many
people were exposed to them. Which is why so many people think these
machines no longer exist
The problem of /how do we get $amount into every employee's bank ac-
count/ with thousands of employees and perhaps a single payroll account
for the company does not sound like a solution waiting for a cluster of
Linux boxes.
That was essentially my point.
Neither is necessarily running a global airline booking application.
Do keep up with the industry, please. E.g. Amadeus - in their own words "WerCOve freed ourselves from legacy mainframe systems and have adapted and evolved our existing IT framework to todayrCOs cloud environment."[0]
We have a range of tools at our disposal, from a Raspberry Pi Pico to an
IBM mainframe, and none of them are suitable for *all* applications.
So you pick the machines that fit the job, of course.
In the middle, a linux *86 style machine is a very useful and cost
effective machine, but not at the spectrum edges.
That "middle" these days is reaching up all the way to 16 sockets (so easily thousands of cores depending one the CPUs used) and 32 TiB of RAM.
E.g. one of the certified platforms for SAP HANA[1] is the Google Cloud x4-1920-32t-metal VM instance: 32 TiB RAM, 1920 vCPUs, running various flavors of e.g. Redhat or SLES Linux.
But then, SAP has always been an epic resource hog and SAP HANA even
more so.
And the usual supercomputers run variants of Linux these days too,
e.g. the top two: Lineshine (China) and El Capitan (USA) both run
custom Linux distributions.
Kind regards,
Alex.
[0] https://amadeus.com/en/blog/articles/five-things-to-know-amadeus-travel-platform
[1] https://www.sap.com/dmc/exp/2014-09-02-hana-hardware/enEN/
On 2026-11-15, The Natural Philosopher <tnp@invalid.invalid> wrote:
On 15/11/2026 12:22, Ian wrote:
Yes, eventually the modern solution wins, but IMO this is the primary
reason "mainframes" still exist.
So today, "mainframe" = "System capable of running software developed
before Linux or Windows was ever conceived" :)
Simply not true
All trad OS and languages run on linux machines of the standard sort -
Fortran - COBOL - RPG...SQL.
It simply is true, and it pays my wages. Yes, of course COBOL et. el. are available on modern OS's, but not the dialects in which all that business logic is encoded. And the business logic is a lot more than a single
COBOL program, it's literally hundreds of them, all knitted together in strange and interesting ways by proprietary job control language scripting, and all the data is in legacy proprietary databases, not nice modern relational ones. Unravelling that is a huge effort, and organisations have spent millions trying and failing to do it, so they're still paying for
the mainframe OS (which is now running on Linux, of course!)
It won't last forever, but it's still doing the job and paying the bills, literally.
On Sat, 14 Nov 2026 00:29:43 -0500
c186282 <c186282@nnada.net> wrote:
On 11/13/26 16:02, Alexander Schreiber wrote:[]
The Natural Philosopher <tnp@invalid.invalid> wrote:
On 13/11/2026 14:40, Peter Flass wrote:
[]
I can't understand (well, I can...) why LiO2 batteries were ever
approved for general use.
There is no such battery. Lithium Ion or LIPO for lithium ion polymer or >> >> LiFePo for Lithium iron polymer
Well, there are Lithium-Air batteries. But while various implementations >> > of them have been demonstrated, AFAIK they are nowhere near
commercialization.
Just the other day I saw on the news that a
house burned (actually exploded) because if a battery fire.
Batteries made badly and sold out of spec.
And/or badly abused during transport, installation, operation. Plus
people with little knowledge of the technology tend to end up with
the more exciting (read: fire-prone) lithium battery chemistries instead >> > of with much safer ones e.g. LiFePo4. It doesn't help that often the
cell chemistry is not explicitly mentioned, e.g. with battery packs
for e-bikes and such.
The best alt these days are sodium batteries ... alreadyWhatever happened to Iron-Air batteries?
commercial, you can buy ones meant to start up automobiles
on Amazon. Not quite the power density, but getting close
and don't explode.
As for 'quality batteries', just no WAY to know what
you're really getting these days.
https://en.wikipedia.org/wiki/Glass_battery
https://en.wikipedia.org/wiki/Solid-state_battery
Lots of "promise", when do we get them in the shops ^w online?
On 15/11/2026 12:43, Johann 'Myrkraverk' Oskarsson wrote:
Bottom line, do we care if we run the Linux on a mainframe?
Nope. But it probably would need to be optimised for it
Mainframes today tend to run just one or two applications and the
operating system is whatever works with those
On Sun, 15 Nov 2026 12:22:30 -0000 (UTC), Ian wrote:
The problem is actually how do we get the *right* amount into every
employee's bank account (or charge the *right* amount of tax, or
give the *right* amount of benefits...), where *right* is defined by
years or decades of ad-hoc business logic and legislation that is
embedded into ancient software (COBOL/JCL et. al.) that can only run
on the "mainframe" it was developed on.
Too often the shiny new solution doesn't get the right answers,
despite being infinitely faster. So what do you do - keep paying
your mainframe supplier and not have any risk, or switch to a modern
solution and risk your job if thousands of people complain that
you've got the sums wrong?
Salaries donrCOt stand still; benefits and deductions donrCOt stand still; taxes donrCOt stand still. Keep running old code, and you find it harder
and harder to keep it up to date with all the new laws and
regulations.
On Sun, 15 Nov 2026 18:04:33 +0800, Johann 'Myrkraverk' Oskarsson wrote:
JSON is probably the most popular today. I personally don't like it for
being wasteful.
I think that record goes to XML. After 9/11 it became painfully apparent
that the various agencies couldn't communicate with each other. The DOJ
set out to fix that and XML was the shiny new technology.
https://en.wikipedia.org/wiki/NIEMOpen
Prior to the fix a return from a NCIC query was a reasonably well
formatted plain text document you could dump to a screen as is. It took a decade before the XML schema was completely adopted. Now the return is a
huge XML document that requires XSLT to extract any human readable information from. Progress.
It doesn't help that the various schemae change. A tag like <Id/> can be almost anything depending on its ancestry. The clients, of course, wanted
the return to be formatted exactly like the plain text returns they had
been getting for years. It kept me in a job, I guess.
There was even a time when IBM actively promoted their mainframes as
HA hosts for running lots of Linux VMs.
On Sat, 14 Nov 2026 00:29:43 -0500
c186282 <c186282@nnada.net> wrote:
On 11/13/26 16:02, Alexander Schreiber wrote:[]
The Natural Philosopher <tnp@invalid.invalid> wrote:
On 13/11/2026 14:40, Peter Flass wrote:
[]
I can't understand (well, I can...) why LiO2 batteries were ever
approved for general use.
There is no such battery. Lithium Ion or LIPO for lithium ion polymer or >>>> LiFePo for Lithium iron polymer
Well, there are Lithium-Air batteries. But while various implementations >>> of them have been demonstrated, AFAIK they are nowhere near
commercialization.
Just the other day I saw on the news that a
house burned (actually exploded) because if a battery fire.
Batteries made badly and sold out of spec.
And/or badly abused during transport, installation, operation. Plus
people with little knowledge of the technology tend to end up with
the more exciting (read: fire-prone) lithium battery chemistries instead >>> of with much safer ones e.g. LiFePo4. It doesn't help that often the
cell chemistry is not explicitly mentioned, e.g. with battery packs
for e-bikes and such.
The best alt these days are sodium batteries ... alreadyWhatever happened to Iron-Air batteries?
commercial, you can buy ones meant to start up automobiles
on Amazon. Not quite the power density, but getting close
and don't explode.
As for 'quality batteries', just no WAY to know what
you're really getting these days.
Or glass ?
https://en.wikipedia.org/wiki/Glass_battery
https://en.wikipedia.org/wiki/Solid-state_battery
Lots of "promise", when do we get them in the shops ^w online?
JSON can be just as complicated. There's no rule that an "id": in a sub object is the same as the "id": in the parent object. The one thing XML
has above JSON is the ability to -- in theory -- unambiguously embed one
XML document in another.
On 11/16/2026 1:07 AM, Thomas Koenig wrote:
Johann 'Myrkraverk' Oskarsson <johann@myrkraverk.invalid> schrieb:
The problem of /how do we get $amount into every employee's bank ac-
count/ with thousands of employees and perhaps a single payroll account
for the company does not sound like a solution waiting for a cluster of
Linux boxes.
SAP moved away from mainframes in the early 1990s with SAP R3, and
they do run payrolls.
And SAP S4/HANA runs on... you guessed it, Linux, exclusively.
It does not even run on mainframes any more - it's Intel (not AMD)
or IBM POWER.
It's not exactly clear from https://en.wikipedia.org/wiki/SAP_R/3 but wouldn't the payroll system still run on a single machine, with perhaps
the database on another, and different parts of the application on even
more nodes? Would that machine qualify as a /mainframe/?
On Sun, 15 Nov 2026 22:40:31 +0100, Alexander Schreiber wrote:
There was even a time when IBM actively promoted their mainframes as
HA hosts for running lots of Linux VMs.
Why not any more?
It's not exactly clear from https://en.wikipedia.org/wiki/SAP_R/3 but
wouldn't the payroll system still run on a single machine, with perhaps
the database on another, and different parts of the application on even
more nodes? Would that machine qualify as a /mainframe/?
Conceptually, SAP is centered around a single database ("single source
of truth"). Physically, this will probably be mirrored, different
parts may be on different machines, ...
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Sun, 15 Nov 2026 22:40:31 +0100, Alexander Schreiber wrote:
There was even a time when IBM actively promoted their mainframes as
HA hosts for running lots of Linux VMs.
Why not any more?
I haven't really paid attention to it, maybe they are still doing it?
No idea.
I mean, technically they are of course correct, but using a mainframe to
run Linux VMs might not be the most cost effective approach ...
Johann 'Myrkraverk' Oskarsson <johann@myrkraverk.invalid> schrieb:
It's not exactly clear from https://en.wikipedia.org/wiki/SAP_R/3 but
wouldn't the payroll system still run on a single machine, with perhaps
the database on another, and different parts of the application on even
more nodes? Would that machine qualify as a /mainframe/?
On 11/16/2026 01:29, Thomas Koenig wrote:
Conceptually, SAP is centered around a single database ("single source
of truth"). Physically, this will probably be mirrored, different
parts may be on different machines, ...
Given what I have observed from a distance of how difficult it has been
for so many global companies to move into the SAP platform, it is clear
that this system base creates a tremendous lock-in, just like the IBM architecture did, even if it may run on both Unix and Windows.
Johann 'Myrkraverk' Oskarsson <johann@myrkraverk.invalid> schrieb:
It's not exactly clear from https://en.wikipedia.org/wiki/SAP_R/3 but
wouldn't the payroll system still run on a single machine, with perhaps
the database on another, and different parts of the application on even
more nodes? Would that machine qualify as a /mainframe/?
On 11/16/2026 01:29, Thomas Koenig wrote:
Conceptually, SAP is centered around a single database ("single source
of truth"). Physically, this will probably be mirrored, different
parts may be on different machines, ...
Given what I have observed from a distance of how difficult it has been
for so many global companies to move into the SAP platform, it is clear
that this system base creates a tremendous lock-in, just like the IBM architecture did, even if it may run on both Unix and Windows. I presume Windows will only take you so far up the tree.
| Sysop: | Amessyroom |
|---|---|
| Location: | Fayetteville, NC |
| Users: | 76 |
| Nodes: | 6 (0 / 6) |
| Uptime: | 56:53:48 |
| Calls: | 1,265 |
| Calls today: | 1 |
| Files: | 1,362 |
| D/L today: |
4 files (9,632K bytes) |
| Messages: | 285,828 |