Ron Johnson writes:
> Dumb question: given your out-of-the-box satisfaction, could it be
> that postgresql.conf hasn't been tweaked?
Here are the modified values:
shared_buffers = 8000
wal_buffers = 80
sort_mem = 32000
effective_cache_size = 400000
random_page_cost = 4
autocommit = false
timezone = UTC
I had run a test with effective_cache_size to high value to see what
would happen. Also adjusted random_page_cost:
random_page_cost effective_cache_size elbow
4 40000 675
.5 40000 592
.1 40000 392
4 1000 30
My conclusion is that random_page_cost should be left alone and
effective_cache_size higher is better.
BTW, the hardware is 2 x 2.4ghz Xeon, 1.2GB, SCSI (linux software
raid) with 10K disks. This is close to the production box. Although
we are planning on adding more memory to production.
Rob
