On 2020-11-26, Jorgen Grahn <grahn+nntp@snipabacken.se> wrote:
On Wed, 2020-11-25, Carlos E.R. wrote:
On 25/11/2020 21.26, Harold Johanssen wrote:...
My intent was, when posing this question, to get feedback on the
performance of different network filesystems with the networking
limitations that I described, from anybody who has had experience with >>>> comparable setups. This is an all-Linux network.
Depends on what the actual load your software exerts. Large files? Small >>> files? Random r/w access? That's the worst.
But (I think) it could also be the scenario where you gain the most by
offloading the actual disk I/O to another machine, which perhaps has
disks with better random access, and more cache.
network speeds are (always?) slower than disk bus access speeds. Thus i
wold expect [ local<-> network<->remote computer <-> disk] to be slower than [local computer <-> disk.]
On 11/26/20 12:22 AM, David Brown wrote:
Channel bundling can help on a good network setup, but it's unlikely to
help if he has cheapo or ancient switches, as he must have, since he is
on 100 Mb - they won't get enough parallel port-to-port transfers to
give significant improvements in throughput. (And if he has hubs,
rather than switches, it certainly won't help.)
It's a lot easier to upgrade a switch than to upgrade the wires in the
walls. One can have a modern switch and old wiring. I'd not be so sure
the person has "ancient" switches.
The odds of encountering someone with a hub is almost nil. I remember
when the first consumer switches hit the market and that was more than
20 years ago.
| Sysop: | Amessyroom |
|---|---|
| Location: | Fayetteville, NC |
| Users: | 59 |
| Nodes: | 6 (0 / 6) |
| Uptime: | 19:26:34 |
| Calls: | 810 |
| Calls today: | 1 |
| Files: | 1,287 |
| D/L today: |
10 files (21,017K bytes) |
| Messages: | 193,989 |