I was hoping to just get a "gut reaction" on some pgbench numbers I have, to
see if I'm in the ballpark.
OS: ScientificLinux 6.3, x86_64
Hardware: 4x real disks (not SSD) behind an LSI 9260 in raid10, Xeon E5-2680
with hyperthreading OFF, 128GB of RAM.
Setup: postgresql 8.4.13, ext4, barriers ON, disk write cache *off*, write-
back enabled on the LSI.
I initialized with sizes of 100, 200, and 400.
I've done some tuning of the postgresql config, but mostly I'm just trying to
find out if I'm in the right ballpark.
I ran pgbench from another (similar) host:
pgbench -h BLAH -c 32 -M prepared -t 100000 -S
I get 95,000 to 100,000 tps.
pgbench -h BLAH -c 32 -M prepared -t 100000
seems to hover around 6,200 tps (size 100) to 13,700 (size 400)
Do these basically sniff right?
(NOTE: with barriers off, I get a slight increase - 10% - in the
read-write test, and a larger *decrease* - 15% - with the read-only
test @ 400. No change @ 100)
--
Jon
