X-Original-To: pgsql-performance-postgresql.org@localhost.postgresql.org Received: from localhost (unknown [200.46.204.144]) by svr1.postgresql.org (Postfix) with ESMTP id 186C83A60DB for ; Tue, 7 Dec 2004 04:17:55 +0000 (GMT) Received: from svr1.postgresql.org ([200.46.204.71]) by localhost (av.hub.org [200.46.204.144]) (amavisd-new, port 10024) with ESMTP id 17928-07 for ; Tue, 7 Dec 2004 04:17:44 +0000 (GMT) Received: from floppy.pyrenet.fr (news.pyrenet.fr [194.250.190.2]) by svr1.postgresql.org (Postfix) with ESMTP id 423783A4DC1 for ; Tue, 7 Dec 2004 04:17:44 +0000 (GMT) Received: by floppy.pyrenet.fr (Postfix, from userid 106) id 5179A31941; Tue, 7 Dec 2004 05:17:43 +0100 (MET) From: Christopher Browne X-Newsgroups: pgsql.performance Subject: Re: scaling beyond 4 processors Date: Mon, 06 Dec 2004 23:16:01 -0500 Organization: cbbrowne Computing Inc Lines: 30 Message-ID: References: <41B4E8B6.6059.1FBEABB@localhost> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Complaints-To: usenet@news.hub.org X-message-flag: Outlook is rather hackable, isn't it? X-Home-Page: http://www.cbbrowne.com/info/ X-Affero: http://svcs.affero.net/rm.php?r=cbbrowne User-Agent: Gnus/5.1007 (Gnus v5.10.7) XEmacs/21.4 (Security Through Obscurity, linux) Cancel-Lock: sha1:eO03lk/ExMEWjRoYl5v6sx6CSQE= To: pgsql-performance@postgresql.org X-Virus-Scanned: by amavisd-new at hub.org X-Spam-Status: No, hits=0.069 tagged_above=0 required=5 tests=AWL X-Spam-Level: X-Archive-Number: 200412/128 X-Sequence-Number: 9503 In the last exciting episode, vogler@cipsoft.com wrote: > Hello everyone! > > Since our current Postgres server, a quad Xeon system, finally can't > keep up with our load anymore we're ready to take the next step. > > So the question is: Has anyone experiences with running Postgres on > systems with more than 4 processors in a production environment? > Which systems and architectures are you using (e.g. IBM xseries, IBM > pseries, HP Proliant, Sun Fire, 8- way Opteron)? How about conflicts > between Postgres' shared memory approach and the NUMA architecture > of most multi-processor machines? The perhaps odd thing is that just about any alternative to quad-Xeon is likely to be _way_ better. There are some context switching problems that lead to it being remarkably poorer than you'd expect. Throw in less-than ideal performance of the PAE memory addressing system and it seems oddly crippled overall. We've been getting pretty good results with IBM pSeries systems; they're expensive, but definitely very fast. Preliminary results with Opterons are also looking very promising. One process seemed about 25x as fast on a 4-way 8GB Opteron as it was on a 4-way 8GB Xeon, albeit with enough differences to make the comparison dangerous. -- wm(X,Y):-write(X),write('@'),write(Y). wm('cbbrowne','gmail.com'). http://www.ntlug.org/~cbbrowne/sgml.html The IETF motto: "Rough consensus *and* working code."