> I notice from your postings on the PGSQL-hackers archives that you have been
> experimenting with PGSQL and Pgbench. I have not been able to get PGSQL to
> scale on multiple processors. Tom Lane says this is due to the
> implementation of spin-locks. Have you made any fixes or additions to PGSQL
> to improve scalability on multiple processors?
No.
> Is there anything else I
> should try?
The only thing I noticed to increase the performance so far is extend
the deadlock_timeout paramterter.
BTW, wal_sync_method parameter seems also impact the performance. I'm
not sure what kind of platform you are using, but from my experience
open_sync is the best for Unix, while fdatasync is good for Solaris.
--
Tatsuo Ishii