Andras Fabian wrote:
> - all fast servers show the COPY process as being in the state Rs ("runnable (on run queue)")
> - on the still slow server, this process is in 9 out of 10 samples in Ds ("uninterruptible sleep (usually IO)")
>
I've run into significant performance regressions in PostgreSQL
performance due to issues with the Linux scheduler before, specifically
when running a single really intensive client program. You might be
seeing something similar here. I wrote a reference link heavy blog
entry about that at
http://notemagnet.blogspot.com/2008/05/pgbench-suffering-with-linux-2623-2626.html
you might find useful, one of the batch scheduler tweaks alluded to
there might improve things. Regression here in newer kernels are the
norm rather than the exception, and given the general lack of quality
control in Ubuntu 10.04 I have avoided any performance testing of it
yet. I was going to give it six months after release before I even
thought about that, in hopes more bugs are squashed, but I'm not
optimistic about this distribution for server use at all right now.
There's more information about using oprofile at
http://wiki.postgresql.org/wiki/Profiling_with_OProfile that might help
you dig into the underlying spot it's stuck at.
--
Greg Smith 2ndQuadrant US Baltimore, MD
PostgreSQL Training, Services and Support
greg@2ndQuadrant.com www.2ndQuadrant.us