From: | Josh Berkus <josh(at)agliodbs(dot)com> |
---|---|
To: | sfpug(at)postgresql(dot)org |
Cc: | Brian Ghidinelli <brian(at)pukkasoft(dot)com> |
Subject: | Re: 8.2 slower than 8.1? |
Date: | 2008-01-03 21:55:43 |
Message-ID: | 200801031355.44373.josh@agliodbs.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | sfpug |
Brian,
You have an n-variable problem here. There are 4 points of difference
between the two machines in addition to the PostgreSQL version. Also,
some comments:
> Prior Now
> -------------------------------------------------------------------
> OS CentOS 4.2 CentOS 5.0
> Filesystem ext3 ext3
> postgresql.conf Stock Stock
A "stock" .conf is not considered terrificially useful for real testing.
> Ram 1 GB 3 GB
> Postgres 8.1.2 8.2.5
> Smart Array 5i (builtin, 16mb) 5302 (128mb cache)
Have you done raw FS performance testing on this array? If the underlying
FS is slow due to driver issues (or something) PG can't do much about it.
> I ran my benchmark script again (see below) and the performance dropped
> 60% at 1 and 2 users and 27% and 17% at 5 and 10 users respectively.
Running pgbench a single time with only one, small, DB size isn't very
useful for determining performance. Small DBs test memory performance,
and it takes large DBs (like s=500) to test disk array peformance. So at
this point the differences in disk array pretty much don't matter in your
test (except for the log disk) and you're letting random factors influence
your comparison.
see: http://www.westnet.com/~gsmith/content/postgresql/pgbench-scaling.htm
--
--Josh
Josh Berkus
PostgreSQL @ Sun
San Francisco
From | Date | Subject | |
---|---|---|---|
Next Message | David Fetter | 2008-01-05 20:33:54 | Using Your Favorite Language Inside Postgres |
Previous Message | Brian Ghidinelli | 2007-12-28 01:43:16 | 8.2 slower than 8.1? |