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 853615671E for ; Tue, 15 Mar 2005 02:23:41 +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 45830-08 for ; Tue, 15 Mar 2005 02:23:30 +0000 (GMT) Received: from sss.pgh.pa.us (sss.pgh.pa.us [66.207.139.130]) by svr1.postgresql.org (Postfix) with ESMTP id 905A55671A for ; Tue, 15 Mar 2005 02:23:27 +0000 (GMT) Received: from sss2.sss.pgh.pa.us (tgl@localhost [127.0.0.1]) by sss.pgh.pa.us (8.13.1/8.13.1) with ESMTP id j2F2NT5X001677; Mon, 14 Mar 2005 21:23:29 -0500 (EST) To: "Greg Sabino Mullane" Cc: pgsql-performance@postgresql.org Subject: Re: cpu_tuple_cost In-reply-to: References: Comments: In-reply-to "Greg Sabino Mullane" message dated "Tue, 15 Mar 2005 02:05:01 +0000" Date: Mon, 14 Mar 2005 21:23:29 -0500 Message-ID: <1676.1110853409@sss.pgh.pa.us> From: Tom Lane X-Virus-Scanned: by amavisd-new at hub.org X-Spam-Status: No, hits=0.007 tagged_above=0 required=5 tests=AWL X-Spam-Level: X-Archive-Number: 200503/291 X-Sequence-Number: 10996 "Greg Sabino Mullane" writes: > On that note, can I raise the idea again of dropping the default > value for random_page_cost in postgresql.conf? I think 4 is too > conservative in this day and age. Certainly the person who will > be negatively impacted by a default drop of 4 to 3 will be the > exception and not the rule. The ones who'd be negatively impacted are the ones we haven't been hearing from ;-). To assume that they aren't out there is a logical fallacy. I still think that 4 is about right for large databases (where "large" is in comparison to available RAM). Also, to the extent that we think these numbers mean anything at all, we should try to keep them matching the physical parameters we think they represent. I think that the "reduce random_page_cost" mantra is not an indication that that parameter is wrong, but that the cost models it feeds into need more work. One thing we *know* is wrong is the costing of nestloop inner indexscans: there needs to be a correction for caching of index blocks across repeated scans. I've looked at this a few times but not come up with anything that seemed convincing. Another thing I've wondered about more than once is if we shouldn't discount fetching of higher-level btree pages on the grounds that they're probably in RAM already, even if the indexscan isn't inside a loop. regards, tom lane