Slow seek in kernels >= 2.6.18

Slow seek in kernels >= 2.6.18

[Anton A Vorobev]

Hi!

A question. We've got an application that runs about 30-40 minutes. 
After kernel upgrade from 2.6.17 branch, it runs about 3 hours.
I've made tests with the help of bonnie++, it shows that random seek 
operations are about 3 times slower in kernels newer than 2.6.18 in 
comparison with 2.6.17. This difference may be seen only during high 
load of a server. When it is in normal state, parameters are rather the 
same.

We use XFS filesystem. Does any one know may it be because of upgrade of 
XFS driver in kernel? Any ideas?

Thanks a lot.

______________________________________________________________
Anton Vorobev
Programátor
Seznam.cz, a.s. Radlická 608/2 150 00 Praha 5

tel.: +420 234 694 706
fax: +420 234 694 115
anton.vorobev@firma.seznam.cz
http://www.seznam.cz

Re: Slow seek in kernels >= 2.6.18

[Bhagi rathi]

Is it the same on ext file-systems?

Can you run the same tests on disk devices directly to rule out scheduling
issues of  i/o requests
by device drivers?  One way I was thinking that we got a kind of cache
pollution or invalidation
thus exposing to do more seeks.  Can you tell the equivalent file-system
operations that your
application in general do? This will give a lot of insight on what could be
happening.

-Bhagi.

On 10/18/07, Anton A Vorobev <anton.vorobev@firma.seznam.cz> wrote:
>
> Hi!
>
> A question. We've got an application that runs about 30-40 minutes.
> After kernel upgrade from 2.6.17 branch, it runs about 3 hours.
> I've made tests with the help of bonnie++, it shows that random seek
> operations are about 3 times slower in kernels newer than 2.6.18 in
> comparison with 2.6.17. This difference may be seen only during high
> load of a server. When it is in normal state, parameters are rather the
> same.
>
> We use XFS filesystem. Does any one know may it be because of upgrade of
> XFS driver in kernel? Any ideas?
>
> Thanks a lot.
>
> ______________________________________________________________
> Anton Vorobev
> Programátor
> Seznam.cz, a.s. Radlická 608/2 150 00 Praha 5
>
> tel.: +420 234 694 706
> fax: +420 234 694 115
> anton.vorobev@firma.seznam.cz
> http://www.seznam.cz
>
>
>
>


[[HTML alternate version deleted]]

Re: {WHAT?} Slow seek in kernels >= 2.6.18

[Peter Grandi]

>>> On Thu, 18 Oct 2007 10:12:28 +0200, Anton A Vorobev
>>> <anton.vorobev@firma.seznam.cz> said:

anton.vorobev> Hi!  A question. We've got an application that
anton.vorobev> runs about 30-40 minutes. After kernel upgrade
anton.vorobev> from 2.6.17 branch, it runs about 3 hours. [
anton.vorobev> ... ]

I'd check the elevator and its parameters more than XFS.

Re: Slow seek in kernels >= 2.6.18

[Josef Sipek]

On Thu, Oct 18, 2007 at 10:12:28AM +0200, Anton A Vorobev wrote:
> Hi!
>
> A question. We've got an application that runs about 30-40 minutes. After 
> kernel upgrade from 2.6.17 branch, it runs about 3 hours.
> I've made tests with the help of bonnie++, it shows that random seek 
> operations are about 3 times slower in kernels newer than 2.6.18 in 
> comparison with 2.6.17. This difference may be seen only during high load 
> of a server. When it is in normal state, parameters are rather the same.

I think that's about when the defaults for few /proc/sys/vm/ vars changed.
One from 10 to 5, and another from 40 to 10. I forget the exact filenames.

Josef 'Jeff' Sipek.

-- 
Once you have their hardware. Never give it back.
(The First Rule of Hardware Acquisition)