On Nov 21, 2007 7:22 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> FWIW, the test cases I was just comparing are entirely CPU-bound ---
> vmstat says there are no disk reads happening at all. Now I only got a
> 3% drop, so that may not be the same effect Guillaume is seeing. But
> the whole thing is a bit upsetting seeing that we thought we'd reduced
> the overhead for short read-only transactions ...
A month or so ago I mentioned to Bruce that we were seeing this
behavior (accidentally) but hadn't had time to focus on it or
determine whether it was a tuning issue.
Basically we're performing the same select-only pgbench test, but with
a varying scale from 1 to 1000. In almost all cases, 8.2.5 is faster
than 8.3 by about 2-5 percent.
The script given to us by a customer was as follows:
for scale in 1 2 5 10 20 30 40 50 75 100 200 400 800 1000; do
echo "------------------------------------------------------------"
echo "SCALE $scale"
dropdb pgbench
createdb pgbench
pgbench -p 5432 -i -s $scale pgbench
psql pgbench -c 'CHECKPOINT'
pgbench -p 5432 -c 8 -t 2500 -S pgbench
pgbench -p 5432 -c 8 -t 2500 -S pgbench
pgbench -p 5432 -c 8 -t 2500 -S pgbench
done
--
Jonah H. Harris, Sr. Software Architect | phone: 732.331.1324
EnterpriseDB Corporation | fax: 732.331.1301
499 Thornall Street, 2nd Floor | jonah.harris@enterprisedb.com
Edison, NJ 08837 | http://www.enterprisedb.com/