Received: from localhost (unknown [200.46.204.184]) by postgresql.org (Postfix) with ESMTP id C03712E0045 for ; Fri, 15 Feb 2008 14:37:33 -0400 (AST) Received: from postgresql.org ([200.46.204.71]) by localhost (mx1.hub.org [200.46.204.184]) (amavisd-maia, port 10024) with ESMTP id 00933-01-6 for ; Fri, 15 Feb 2008 14:37:28 -0400 (AST) Received: from smtp160.iad.emailsrvr.com (smtp160.iad.emailsrvr.com [207.97.245.160]) by postgresql.org (Postfix) with ESMTP id 4565A2E0038 for ; Fri, 15 Feb 2008 14:37:30 -0400 (AST) Received: from relay6.relay.iad.emailsrvr.com (localhost [127.0.0.1]) by relay6.relay.iad.emailsrvr.com (SMTP Server) with ESMTP id A1C34709732; Fri, 15 Feb 2008 13:37:29 -0500 (EST) Received: by relay6.relay.iad.emailsrvr.com (Authenticated sender: erik-AT-myemma.com) with ESMTP id 6126170AD16; Fri, 15 Feb 2008 13:37:29 -0500 (EST) In-Reply-To: <200802151006.11858.josh@agliodbs.com> References: <20080215123528.GA16532@hyperion.scode.org> <20080215135846.GA17327@hyperion.scode.org> <200802151006.11858.josh@agliodbs.com> Mime-Version: 1.0 (Apple Message framework v752.3) Content-Type: text/plain; charset=ISO-8859-1; delsp=yes; format=flowed Message-Id: <72BE31DF-ABAF-459B-99A3-4E575EC3E08A@myemma.com> Cc: pgsql-performance@postgresql.org, Greg Smith , Peter Schuller Content-Transfer-Encoding: quoted-printable From: Erik Jones Subject: Re: shared_buffers in 8.3 w/ lots of RAM on dedicated PG machine Date: Fri, 15 Feb 2008 12:37:10 -0600 To: Josh Berkus X-Mailer: Apple Mail (2.752.3) X-Virus-Scanned: Maia Mailguard 1.0.1 X-Archive-Number: 200802/183 X-Sequence-Number: 28725 On Feb 15, 2008, at 12:06 PM, Josh Berkus wrote: > On Friday 15 February 2008 06:29, Greg Smith wrote: >> PostgreSQL only uses direct I/O for writing to the WAL; everything =20= >> else >> goes through the regular OS buffer cache unless you force it to do >> otherwise at the OS level (like some Solaris setups do with >> forcedirectio). > > Also, note that even when direct I/O is available, most users and =20 > benchmark > tests have reported that having PostgreSQL "take over" the entire =20 > cache is > not a net performance gain. I believe this is mostly because our I/=20= > O and > caching code aren't designed for this kind of operation. > > I believe that MyEmma had a different experience on their workload, =20= > though. Actually, while we did have shared_buffers set to 4G on an 8G system =20 when we were running with forcedirectio, the decision to even run =20 with forcedirectio was a temporary until we were able (welll, forced =20 to) migrate to a new system with a sane drive configuration. The old =20= set up was done horribly by a sysadmin who's no longer with us who =20 set us up with a RAID5 array with both the data and xlogs both =20 mirrored across all of the disks with no spares. So, I wouldn't =20 consider the numbers I was seeing then a reliable expectation as that =20= system was nowhere close to ideal. We've seen much more sane and =20 consistent numbers on a more normal setup, i.e. without forcedirectio =20= and with <=3D 25% system memory. Erik Jones DBA | Emma=AE erik@myemma.com 800.595.4401 or 615.292.5888 615.292.0777 (fax) Emma helps organizations everywhere communicate & market in style. Visit us online at http://www.myemma.com