On Sat, 24 Nov 2007, Peter Bauer wrote:
> top shows that the CPUs are at least 80% idle most of the time so i
> think there is an I/O bottleneck.
top also shows that you're never waiting for I/O which is usually evidence
there isn't an I/O bottleneck. You passed along most of the right data,
but some useful additional things to know are:
-Actual brand/model of SCSI controller
-Operating system
-What time interval the vmstat and iostat information you gave were
produced at.
I agree with Scott that checkpoints should be considered as a possibility
here. I'd suggest you set checkpoint_warning to a high value so you get a
note in the logs every time one happens, then see if those happen at the
same time as your high load average. More on that topic and how to adjust
the background writer if that proves to be the cause of your slowdown is
at http://www.westnet.com/~gsmith/content/postgresql/chkp-bgw-83.htm
--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD
