Re: lazy vxid locks, v1

On 06/13/2011 07:55 AM, Stefan Kaltenbrunner wrote:
> all those tests are done with pgbench running on the same box - which
> has a noticable impact on the results because pgbench is using ~1 core
> per 8 cores of the backend tested in cpu resoures - though I don't think
> it causes any changes in the results that would show the performance
> behaviour in a different light.
>

Yeah, this used to make a much bigger difference, but nowadays it's not
so important. So long as you have enough cores that you can spare a
chunk of them to drive the test with, and you crank "-j" up to a lot,
there doesn't seem to be much of an advantage to moving the clients to a
remote system now. You end up trading off CPU time for everything going
through the network stack, which adds yet another set of uncertainty to
the whole thing anyway.

I'm glad to see so many people have jumped onto doing these SELECT-only
tests now. The performance farm idea I've been working on runs a test
just like what's proven useful here. I'd suggested that because it's
been really sensitive to changes in locking and buffer management for me.

--
Greg Smith 2ndQuadrant US greg(at)2ndQuadrant(dot)com Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support www.2ndQuadrant.us