Received: from localhost (unknown [200.46.204.187]) by postgresql.org (Postfix) with ESMTP id CD0A82E30D9 for ; Tue, 8 Jan 2008 12:49:09 -0400 (AST) Received: from postgresql.org ([200.46.204.71]) by localhost (mx1.hub.org [200.46.204.187]) (amavisd-maia, port 10024) with ESMTP id 74171-07 for ; Tue, 8 Jan 2008 12:48:57 -0400 (AST) Received: from lists.commandprompt.com (host-159.commandprompt.net [207.173.203.159]) by postgresql.org (Postfix) with ESMTP id F05112E30D5 for ; Tue, 8 Jan 2008 12:49:03 -0400 (AST) Received: from [192.168.1.226] (or-69-34-217-90.sta.embarqhsd.net [69.34.217.90]) (authenticated bits=0) by lists.commandprompt.com (8.13.8/8.13.8) with ESMTP id m08Gn97d021715 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO); Tue, 8 Jan 2008 08:49:09 -0800 Message-ID: <4783A984.10602@commandprompt.com> Date: Tue, 08 Jan 2008 08:49:08 -0800 From: "Joshua D. Drake" Organization: Command Prompt, Inc. User-Agent: Thunderbird 2.0.0.9 (Windows/20071031) MIME-Version: 1.0 To: Hans Guijt CC: pgsql-admin@postgresql.org Subject: Re: Less intrusive ways to cluster? References: <85978993146E1A4AA9CC4B2945CAFC940233AD76@exch-be.terma.com> In-Reply-To: <85978993146E1A4AA9CC4B2945CAFC940233AD76@exch-be.terma.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit X-Greylist: Sender succeeded SMTP AUTH, not delayed by milter-greylist-4.0 (lists.commandprompt.com [207.173.203.159]); Tue, 08 Jan 2008 08:49:09 -0800 (PST) X-Virus-Scanned: Maia Mailguard 1.0.1 X-Archive-Number: 200801/84 X-Sequence-Number: 27858 Hans Guijt wrote: > I have a fairly large table that keeps track of data measured by our > system. The data is in the form of BLOBs, and is only queried in order > of timestamp. Because we are measuring from multiple devices, the data > does not necessarily arrive in the database in correct temporal order, > and in fact it is very well possible for corrections to some BLOBs to be > entered after measurement. As a result, there is considerable churn in > the table - but only at the very end, in the last hour or so of data. > Older data is normally left alone. > Alternatively, is there some way to do partial clustering? Since 99% of > my data set will already be properly clustered, except for the last 24 > hours or so of data, just clustering that last bit (which is trivial by > comparison) would already help a great deal. However, I'm not sure how > to achieve this. Why not break it up into a partitioning scheme? Joshua D. Drake >