On Mon, Jan 31, 2005 at 01:31:23PM +0000, Abdul-Wahid Paterson wrote:
> What exactly are the Xeon spinlock issues? Do you have a reference for
> more information. Just that I am thinking about which hardware to
> purchase now for my new server array.

You should have a look in the pgsql-performance and pgsql-hackers
archives.  Look for "spinlock" over the past 12 months or so.  There
has been a great deal of discussion about it.

The short version is that PostgreSQL has spinlock issues on more or
less all platforms under concurrent loads; that SMP makes it worse;
and that Xeons, in particular, are especially bad for this case. 
There's something about the hardware which seem to exacerbate the
problem; details are found in the above-mentioned lists, and I'm
leery of mentioning particulars off the top of my head, because I'll
get it wrong.  In testing, Afilias (my employer) has had some
especially good results with Opterons as an alternative (for our
purposes, we saw something better than 10x improvement, but that was
using our own application as a benchmark).  We've also been
pleasantly surprised by the comparitive performance on AIX versus
Solaris, in case larger systems are your target (we're using IBM
RS6000 P650s for some applications.  They're real fast, but AIX is
mighty strange if you're not used to it).  This is not a
recommendation, your milage may vary, disclaimer disclaimer &c.

A

-- 
Andrew Sullivan  | ajs at crankycanuck.ca
The plural of anecdote is not data.
		--Roger Brinner