Обсуждение: context-switching issue on Xeon

FWIW, I'm seeing this with a client at the moment. 40-60k CS per second
on Dual 3.2GHz.

There are plenty of other issues we're dealing with, but this is
obviously
disconcerting...


On 19 Jul 2005, at 19:23, Sailer, Denis (YBUSA-CDR) wrote:

> The thread below has the test case that we were able to use to
> reproduce the issue.
>
>
> http://archives.postgresql.org/pgsql-performance/2004-04/msg00280.php
>
>
> The last messages on this subject are from April of 2005.  Has
> there been any successful ways to significantly reduce the impact
> this has to multi-processing?  I haven’t been able to find anything
> showing a resolution of some kind.
>
>
> We are seeing this on two of our machines:
>
>
> Quad 3.0 GHz XEON with 3GB of memory running PG 7.4.3 with SuSE
> kernel 2.4
>
>
> Dual 2.8 GHz XEON with 2GB of memory running PG 8.0.0 with SuSE
> kernel 2.4
>
>
>
>
>

"Sailer, Denis (YBUSA-CDR)" <Denis.Sailer@Yellowbook.com> writes:
> http://archives.postgresql.org/pgsql-performance/2004-04/msg00280.php

> The last messages on this subject are from April of 2005.  Has there
> been any successful ways to significantly reduce the impact this has to
> multi-processing?

CVS tip should largely fix the problem as far as buffer manager
contention goes.

> I haven't been able to find anything showing a
> resolution of some kind.

Look at the Feb/March threads concerning buffer manager rewrite, clock
sweep, etc ... eg
http://archives.postgresql.org/pgsql-patches/2005-03/msg00015.php

            regards, tom lane