I am trying to recall how to tweak a kernel for heavy traffic.
I am trying to recall how to tweak a kernel for heavy traffic.
What is the issue and where is the bottleneck?
Did you check on which part a software is waiting?
----
Gru|f
Marco
Junk-Mail bitte an trashcan@stinkedores.dorfdsl.de
On my nginx system, Large blog posts choke.
Since taking a full feed from NNTP, my posting takes second to post.
Define "full feed". Do you a mean a full binary feed? If so, no wonder, for a true full feed you need 4 or 5 servers with very fast storage and at least 10Gbps of bandwidth to keep up. If you mean a full text feed, that will do nothing to any modern computer in terms of impact performance.
I've never had to tune FreeBSD for web, database, SMTP, or NNTP. I did mess around with a lot of suggestions for making NFS and Samba/SMB faster, but none
of it worked. Unsure the technical reason, but Linux beats the pants off FreeBSD for NFS and SMB performance. Could never get more than ~55Mbps throughput with FreeBSD, but can get near wire-speed from Linux. If anyone happens to read this and has suggestions, I'm all ears.
On Jul 16, 2026 at 5:18:51rC>PM CDT, "The Doctor" <The Doctor> wrote:
On my nginx system, Large blog posts choke.
Since taking a full feed from NNTP, my posting takes second to post.
Define "full feed". Do you a mean a full binary feed? If so, no wonder, for a >true full feed you need 4 or 5 servers with very fast storage and at least >10Gbps of bandwidth to keep up. If you mean a full text feed, that will do >nothing to any modern computer in terms of impact performance.
I've never had to tune FreeBSD for web, database, SMTP, or NNTP. I did mess >around with a lot of suggestions for making NFS and Samba/SMB faster, but none >of it worked. Unsure the technical reason, but Linux beats the pants off >FreeBSD for NFS and SMB performance. Could never get more than ~55Mbps >throughput with FreeBSD, but can get near wire-speed from Linux. If anyone >happens to read this and has suggestions, I'm all ears.
Il 17/07/2026 00:42, Jesse Rehmer ha scritto:
Define "full feed". Do you a mean a full binary feed? If so, no wonder, for a
true full feed you need 4 or 5 servers with very fast storage and at least >> 10Gbps of bandwidth to keep up. If you mean a full text feed, that will do >> nothing to any modern computer in terms of impact performance.
Yep. He asked me for a binary feed a few days ago, and since then I
think his server has been "hot". I warned him that this might happen (I
have four frontend servers and five upstream ones, so I'm not having any >problems) ^_^"
I've never had to tune FreeBSD for web, database, SMTP, or NNTP. I did mess >> around with a lot of suggestions for making NFS and Samba/SMB faster, but none
of it worked. Unsure the technical reason, but Linux beats the pants off
FreeBSD for NFS and SMB performance. Could never get more than ~55Mbps
throughput with FreeBSD, but can get near wire-speed from Linux. If anyone >> happens to read this and has suggestions, I'm all ears.
I see it depends on the hardware (chipset). And if you use ramfs instead
of tmpfs (but it's a matter of taste). And in any case, some drivers
used in the Linux kernel are undoubtedly better than those of *BSD >(especially now that BSD has eliminated the entire GNU part to be
completely in line with the BSD license and not the GPL).
----
Ivo Gandolfo
In article <113bmo6$dtb$1@omnios-inn.blueworldhosting.com>,
Jesse Rehmer <jesse.rehmer@blueworldhosting.com> wrote:
I've never had to tune FreeBSD for web, database, SMTP, or NNTP. I did mess >>around with a lot of suggestions for making NFS and Samba/SMB faster, but none
of it worked. Unsure the technical reason, but Linux beats the pants off >>FreeBSD for NFS and SMB performance. Could never get more than ~55Mbps >>throughput with FreeBSD, but can get near wire-speed from Linux. If anyone >>happens to read this and has suggestions, I'm all ears.
I recall tuning my BSD/OS server. Same method should apply.
On Fri, 17 Jul 2026 03:18:23 -0000 (UTC), The Doctor wrote:
In article <113bmo6$dtb$1@omnios-inn.blueworldhosting.com>,
Jesse Rehmer <jesse.rehmer@blueworldhosting.com> wrote:
I've never had to tune FreeBSD for web, database, SMTP, or NNTP. I did mess >>>around with a lot of suggestions for making NFS and Samba/SMB faster, but none
of it worked. Unsure the technical reason, but Linux beats the pants off >>>FreeBSD for NFS and SMB performance. Could never get more than ~55Mbps >>>throughput with FreeBSD, but can get near wire-speed from Linux. If anyone >>>happens to read this and has suggestions, I'm all ears.
I recall tuning my BSD/OS server. Same method should apply.
It's been a long time since I adjusted kernel knobs, but
you may want to look into one or more of
% sysctl -a | grep -E recv | more
net.local.stream.recvspace: 65536
net.local.dgram.recvspace: 16384
net.local.seqpacket.recvspace: 65536
net.inet.igmp.recvifkludge: 1
net.inet.tcp.recvspace: 65536
net.inet.tcp.recvbuf_max: 8388608
net.inet.tcp.recvbuf_auto: 1
net.inet.udp.recvspace: 42080
net.inet.raw.recvspace: 9216
net.key.recvspace: 8192
net.netlink.recvspace: 65536
net.wlan.recv_bar: 1
net.rtsock.recvspace: 8192
There may also be a few NIC specific tunable to look
at. For I have re0, so 'sysctl -a | grep dev.re'.
This NIC's driver seems to not have any relevant knobs
to tweak.
--
steve
In article <113atgf$22len$1@dont-email.me>, Marco Moock <mm@dorfdsl.de> wrote:
Am 16.07.26 um 15:57 schrieb The Doctor:
I am trying to recall how to tweak a kernel for heavy traffic.
What is the issue and where is the bottleneck?
Did you check on which part a software is waiting?
On my nginx system, Large blog posts choke.
Since taking a full feed from NNTP, my posting takes second to post.
In article <113atgf$22len$1@dont-email.me>, Marco Moock<mm@dorfdsl.de> wrote:
Am 16.07.26 um 15:57 schrieb The Doctor:
I am trying to recall how to tweak a kernel for heavy traffic.
What is the issue and where is the bottleneck?
Did you check on which part a software is waiting?
On my nginx system, Large blog posts choke.
Since taking a full feed from NNTP, my posting takes second to post.
Then check CPU load, RAM usage, disk usage.
Some applications maybe need to wait for one of them.
----
Gru|f
Marco
Junk-Mail bitte an trashcan@stinkedores.dorfdsl.de
In article <113f4mh$3bq75$2@dont-email.me>, Marco Moock <mm@dorfdsl.de> wrote:
Am 17.07.26 um 00:18 schrieb The Doctor:
In article <113atgf$22len$1@dont-email.me>, Marco Moock<mm@dorfdsl.de> wrote:
Am 16.07.26 um 15:57 schrieb The Doctor:
I am trying to recall how to tweak a kernel for heavy traffic.
What is the issue and where is the bottleneck?
Did you check on which part a software is waiting?
On my nginx system, Large blog posts choke.
Since taking a full feed from NNTP, my posting takes second to post.
Then check CPU load, RAM usage, disk usage.
Some applications maybe need to wait for one of them.
Ever read System or Performance Tuning, an O'Reilly Book?
In article <113f4mh$3bq75$2@dont-email.me>, Marco Moock<mm@dorfdsl.de> wrote:
Am 17.07.26 um 00:18 schrieb The Doctor:
In article <113atgf$22len$1@dont-email.me>, Marco Moock<mm@dorfdsl.de> wrote:
Am 16.07.26 um 15:57 schrieb The Doctor:
I am trying to recall how to tweak a kernel for heavy traffic.
What is the issue and where is the bottleneck?
Did you check on which part a software is waiting?
On my nginx system, Large blog posts choke.
Since taking a full feed from NNTP, my posting takes second to post.
Then check CPU load, RAM usage, disk usage.
Some applications maybe need to wait for one of them.
Ever read System or Performance Tuning, an O'Reilly Book?
No, I didn't.
Although, before you can start tuning, you need to figure out the >bottleneck.
----
Gru|f
Marco
Junk-Mail bitte an trashcan@stinkedores.dorfdsl.de
take "man tuning" as a start.
I've never had to tune FreeBSD for web, database, SMTP, or NNTP. I did mess >around with a lot of suggestions for making NFS and Samba/SMB faster, but none >of it worked. Unsure the technical reason, but Linux beats the pants off >FreeBSD for NFS and SMB performance. Could never get more than ~55Mbps >throughput with FreeBSD, but can get near wire-speed from Linux. If anyone >happens to read this and has suggestions, I'm all ears.
In article <113bmo6$dtb$1@omnios-inn.blueworldhosting.com>,
Jesse Rehmer <jesse.rehmer@blueworldhosting.com> wrote:
I've never had to tune FreeBSD for web, database, SMTP, or NNTP. I did mess >> around with a lot of suggestions for making NFS and Samba/SMB faster, but none
of it worked. Unsure the technical reason, but Linux beats the pants off
FreeBSD for NFS and SMB performance. Could never get more than ~55Mbps
throughput with FreeBSD, but can get near wire-speed from Linux. If anyone >> happens to read this and has suggestions, I'm all ears.
Did you set MTU 9000?
I believe the FreeBSD installer will default to 1500.
Am 19.07.26 um 18:13 schrieb John D Groenveld:
In article <113bmo6$dtb$1@omnios-inn.blueworldhosting.com>,but none
Jesse Rehmer <jesse.rehmer@blueworldhosting.com> wrote:
I've never had to tune FreeBSD for web, database, SMTP, or NNTP. I did mess >>> around with a lot of suggestions for making NFS and Samba/SMB faster,
of it worked. Unsure the technical reason, but Linux beats the pants off >>> FreeBSD for NFS and SMB performance. Could never get more than ~55Mbps
throughput with FreeBSD, but can get near wire-speed from Linux. If anyone >>> happens to read this and has suggestions, I'm all ears.
Did you set MTU 9000?
I believe the FreeBSD installer will default to 1500.
Which is reasonable, as the switches and routers needs to be able to
process Jumbo Frames - otherwise the most likely drop the frames.
----
Gru|f
Marco
Spam bitte an abfalleimer2001@stinkedores.dorfdsl.de
process Jumbo Frames - otherwise the most likely drop the frames.
In article <113j5vj$364d$2@solani.org>, Marco Moock <mm@dorfdsl.de> wrote: >>Which is reasonable, as the switches and routers needs to be able to >>process Jumbo Frames - otherwise the most likely drop the frames.
My WAG is Jesse's network supports jumbo frames and the Linux system
he used to benchmark NFS and CIFS was configured with an installer
that correctly guessed the same.
Are there a lot of 1Gbps and higher wired networks that can't handle
MTU 9000?
John
groenveld@acm.org
In article <113j5vj$364d$2@solani.org>, Marco Moock <mm@dorfdsl.de> wrote: >>Which is reasonable, as the switches and routers needs to be able to >>process Jumbo Frames - otherwise the most likely drop the frames.
My WAG is Jesse's network supports jumbo frames and the Linux system
he used to benchmark NFS and CIFS was configured with an installer
that correctly guessed the same.
Are there a lot of 1Gbps and higher wired networks that can't handle
MTU 9000?
I use to do this in BSD/OS
and since 10 years ago, the default Freebsd kernel, until now.
HOw can I tweak the GERNEIC kernel to handle heavy traffic?
Did you set MTU 9000?
I believe the FreeBSD installer will default to 1500.
On 7/16/2026 9:57 PM, The Doctor wrote:
I use to do this in BSD/OS
and since 10 years ago, the default Freebsd kernel, until now.
HOw can I tweak the GERNEIC kernel to handle heavy traffic?
You don't usually directly manipulate
kernel source codes. Are you sure
about what you are trying to do?
FreeBSD is famous for security.
----
@~@ Simplicity is Beauty! Remain silent! Drink, Blink, Stretch!
/ v \ May the Force and farces be with you! Live long and prosper!!
/( _ )\ https://sites.google.com/site/changmw/
^ ^ https://github.com/changmw/changmw
The game is afoot... Meow...
| Sysop: | Amessyroom |
|---|---|
| Location: | Fayetteville, NC |
| Users: | 74 |
| Nodes: | 6 (0 / 6) |
| Uptime: | 50:10:52 |
| Calls: | 1,100 |
| Files: | 1,339 |
| Messages: | 275,859 |