2.4.21 on SMP has extra slow context switch

2.4.21 on SMP has extra slow context switch

[devik]

Hello,

I upgraded one SMP box (Soyo MB, 2x PII/350, 500MB) from
2.4.8 (really) to 2.4.21 last week. Users start to complaint
that it is slow.
I run lmbench's context switch measurer and it's result is:
"size=0k ovr=5.41
2 169.26
3 111.28
4 83.99

While on very similar UP system (the same CPU & kernel) it is:
"size=0k ovr=3.50
2 1.80
3 2.08
4 2.89

Have someone even idea what is going on ? The systems seems to be
fairly stable but slooow.
I plan to reboot with single CPU at evening and/or upgrade to .22,
but the numbers above are weird.

thanks,
-------------------------------
    Martin Devera aka devik
Linux kernel QoS/HTB maintainer
  http://luxik.cdi.cz/~devik/

Re: 2.4.21 on SMP has extra slow context switch

[devik]

New informations. I tested more kernels.
2.4.21 and 22 with 2 CPUs are slow as hell. When I reboot
with maxcpus=1 then all is suddenly ok.
Also floppy drive was not detected in 2 CPU config while
with maxcpus=1 it is.
I placed lmbench data and dmesg output to
http://luxik.cdi.cz/~devik/tmp/slow-smp/ in case someone
knows what to do.
I can do some other tests if you want me to do them.

thanks,
devik

On Wed, 27 Aug 2003, devik wrote:

> Hello,
>
> I upgraded one SMP box (Soyo MB, 2x PII/350, 500MB) from
> 2.4.8 (really) to 2.4.21 last week. Users start to complaint
> that it is slow.
> I run lmbench's context switch measurer and it's result is:
> "size=0k ovr=5.41
> 2 169.26
> 3 111.28
> 4 83.99
>
> While on very similar UP system (the same CPU & kernel) it is:
> "size=0k ovr=3.50
> 2 1.80
> 3 2.08
> 4 2.89
>
> Have someone even idea what is going on ? The systems seems to be
> fairly stable but slooow.
> I plan to reboot with single CPU at evening and/or upgrade to .22,
> but the numbers above are weird.
>
> thanks,
> -------------------------------
>     Martin Devera aka devik
> Linux kernel QoS/HTB maintainer
>   http://luxik.cdi.cz/~devik/
>
>