James,
> The box at my disposal has 64 cores (Power7, 3.6GHz) and 512Gb RAM.
> Disk is currently configured as 6 x 900Gb SAS drives in a RAID 5
> configuration.
You'll want to reconfigure the IO before doing any benchmarks, or you'll
be just measuring the speed of the RAID5.
> So - please let me have your thoughts. If anyone knows of someone who
> would be interested in this please pass this message on (or let me
> know where to post it onto - my job is to be an advocate for
> EnterpriseDB so this seemed like a good mailing list to start with
> but I am aware there are many others). Any suggestions of what to
> benchmark - and I'm looking especially for something that will
> provide a meaningful comparison against other databases (e.g. Oracle)
> where possible. But equally if anyone needs anything testing on a
> Power architecture service with lots of cores and RAM send me the
> details and I'll do all I can to help the community.
So, the main problem is that we don't really have benchmarks which can
max out that kind of hardware. DBT2 will hit limits on how fast the
program can run before the database is saturated, and setting up the kit
for Spec would be a very involved project.
We could do something with having multiple pgbench clients hitting the
DB at once, but that wouldn't be comparable to anything else in the world.
--
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com