Performance degradation on 2.6.25 kernels for http servers

Performance degradation on 2.6.25 kernels for http servers

[Ben Greear]

We noticed something strange today.

We have a test that launches 10 http clients that request a small
URL over and over again.  Keep-alive is enabled on clients and
apache.

When using a 2.6.24 or earlier kernel on the web server, we get around
1200 urls per second completed on each of the 10 clients.  With 2.6.25 kernel
we get around 1000 urls per second completed.  In other words, about
a 20% degradation in connections per second.

We see the same affect when we use standard Fedora kernels as when
we use our own patched kernels.  We see similar degradations when using
FC5 and FC8, and 32-bit and 64-bit installs.  We see similar degradations
on different hardware systems.  We have only tried our modified kernels
and standard Fedora kernels at this time (no un-modified kernels from kernel.org).

It could still be something we are doing wrong, but if anyone
else has a similar test case, I'm curious to know if you see
similar degradations.

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: Performance degradation on 2.6.25 kernels for http servers

[Tobias Koeck]

Do you have the same configuration?

2008/7/23 Ben Greear <greearb@candelatech.com>:
> We noticed something strange today.
>
> We have a test that launches 10 http clients that request a small
> URL over and over again.  Keep-alive is enabled on clients and
> apache.
>
> When using a 2.6.24 or earlier kernel on the web server, we get around
> 1200 urls per second completed on each of the 10 clients.  With 2.6.25
> kernel
> we get around 1000 urls per second completed.  In other words, about
> a 20% degradation in connections per second.
>
> We see the same affect when we use standard Fedora kernels as when
> we use our own patched kernels.  We see similar degradations when using
> FC5 and FC8, and 32-bit and 64-bit installs.  We see similar degradations
> on different hardware systems.  We have only tried our modified kernels
> and standard Fedora kernels at this time (no un-modified kernels from
> kernel.org).
>
> It could still be something we are doing wrong, but if anyone
> else has a similar test case, I'm curious to know if you see
> similar degradations.
>
> Thanks,
> Ben
>
> --
> Ben Greear <greearb@candelatech.com>
> Candela Technologies Inc  http://www.candelatech.com
>
> --
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

Re: Performance degradation on 2.6.25 kernels for http servers

[Ben Greear]

Tobias Koeck wrote:
> Do you have the same configuration?

I'm not sure I understand the question.

We were booting into one kernel, running the test, and
then rebooting with a different kernel.  There were no other
changes to any config files or settings that I am aware of.

I'll try a git-bisect when I get a chance...

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com