X-Original-To: pgsql-performance-postgresql.org@localhost.postgresql.org Received: from localhost (unknown [64.117.224.130]) by svr1.postgresql.org (Postfix) with ESMTP id 6A7EDD1B554 for ; Mon, 6 Oct 2003 18:17:49 +0000 (GMT) Received: from svr1.postgresql.org ([64.117.224.193]) by localhost (neptune.hub.org [64.117.224.130]) (amavisd-new, port 10024) with ESMTP id 15562-10 for ; Mon, 6 Oct 2003 15:17:00 -0300 (ADT) Received: from bob.samurai.com (bob.samurai.com [205.207.28.75]) by svr1.postgresql.org (Postfix) with ESMTP id B691ED1B54D for ; Mon, 6 Oct 2003 15:14:50 -0300 (ADT) Received: from 6-allhosts (d226-85-166.home.cgocable.net [24.226.85.166]) by bob.samurai.com (Postfix) with ESMTP id 48E071F24; Mon, 6 Oct 2003 14:14:34 -0400 (EDT) Subject: Re: reindex/vacuum locking/performance? From: Neil Conway To: Tom Lane Cc: Andrew Sullivan , PostgreSQL Performance In-Reply-To: <1065397835.471.17.camel@tokyo> References: <60d6de9fah.fsf@dev6.int.libertyrms.info> <20031004152241.GL580@libertyrms.info> <1065396767.466.12.camel@tokyo> <15813.1065397428@sss.pgh.pa.us> <1065397835.471.17.camel@tokyo> Content-Type: text/plain Message-Id: <1065464069.473.31.camel@tokyo> Mime-Version: 1.0 X-Mailer: Ximian Evolution 1.4.5 Date: Mon, 06 Oct 2003 14:14:29 -0400 Content-Transfer-Encoding: 7bit X-Virus-Scanned: by amavisd-new at postgresql.org X-Archive-Number: 200310/184 X-Sequence-Number: 3932 On Sun, 2003-10-05 at 19:50, Neil Conway wrote: > On Sun, 2003-10-05 at 19:43, Tom Lane wrote: > > This would be relatively easy to fix as far as our own buffering is > > concerned, but the thing that's needed to make it really useful is > > to prevent caching of seqscan-read pages in the kernel disk buffers. > For the non-portable way of doing this, are you referring to O_DIRECT? I was hoping you'd reply to this, Tom -- you were referring to O_DIRECT, right? (If you were referring to O_DIRECT, I wanted to add that I wouldn't be surprised if using O_DIRECT on many kernels reduces or eliminates any readahead the OS will be doing on the sequential read, so the net result may actually be a loss for a typical seqscan.) -Neil