On Mon, Oct 06, 2003 at 08:07:27AM -0400, Jeff wrote:
> I strongly agree here only if we can prove there is a benefit.
There's plenty of academic work which purports to show that LRU is
far from the best choice. Just in principle, it seems obvious that a
single-case seqscan-type operation (such as vacuum does) is a good
way to lose your cache for no real gain.
> I'd be interested in tinkering with this, but I'm more interested at the
> moment of why (with proof, not antecdotal) Solaris is so much slower than
> Linux and what we cna do about this. We're looking to move a rather large
> Informix db to PG and ops has reservations about ditching Sun hardware.
Interestingly, we're contemplating ditching Solaris because of the
terrible reliability we're getting from the hardware.
You can use truss to find some of the problems on Solaris. The
open() syscall takes forever when you don't hit the Postgres shared
buffers (even if you can be sure the page is in filesystem buffers --
we could demonstrate it on a 1 G database on a machine with 10 G of
RAM). I've heard grumblings about spinlocks on Solaris which might
explain this problem. I certainly notice that performance gets
geometrically worse when you add a few hundred extra connections.
A
--
----
Andrew Sullivan 204-4141 Yonge Street
Afilias Canada Toronto, Ontario Canada
<andrew@libertyrms.info> M2P 2A8
+1 416 646 3304 x110