From pgsql-performance-owner@postgresql.org Tue Sep 10 02:40:47 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 6DA40475C8B for ; Tue, 10 Sep 2002 02:40:46 -0400 (EDT) Received: from ns1.officenet.no (ns1.officenet.no [193.212.174.3]) by postgresql.org (Postfix) with ESMTP id B8F35475C45 for ; Tue, 10 Sep 2002 02:40:45 -0400 (EDT) Received: from [193.212.174.117] (helo=jeb.officenet.no) by ns1.officenet.no with esmtp (Exim 3.36 #3) id 17oehY-0000Yq-00 for pgsql-performance@postgresql.org; Tue, 10 Sep 2002 08:40:44 +0200 Content-Type: text/plain; charset="us-ascii" From: Andreas Joseph Krogh Organization: OfficeNet AS To: pgsql-performance@postgresql.org Subject: subscribe Date: Tue, 10 Sep 2002 08:40:44 +0200 User-Agent: KMail/1.4.3 MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Message-Id: <200209100840.44400.andreak@officenet.no> X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1 X-Sequence-Number: 1 From pgsql-performance-owner@postgresql.org Tue Sep 10 08:31:09 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 7769A4761AF for ; Tue, 10 Sep 2002 08:31:08 -0400 (EDT) Received: from dune.aardvarkmedia.co.uk (unknown [195.224.39.132]) by postgresql.org (Postfix) with ESMTP id 97E5B476092 for ; Tue, 10 Sep 2002 08:31:07 -0400 (EDT) Received: from [192.168.168.13] (helo=Gavin) by dune.aardvarkmedia.co.uk with smtp (Exim 3.33 #1) id 17okAe-0006ek-00 for pgsql-performance@postgresql.org; Tue, 10 Sep 2002 13:31:08 +0100 From: "Gavin Love" To: Subject: subscribe Date: Tue, 10 Sep 2002 13:31:10 +0100 Message-ID: MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit X-Priority: 3 (Normal) X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook IMO, Build 9.0.2416 (9.0.2911.0) X-MimeOLE: Produced By Microsoft MimeOLE V5.50.4522.1200 Importance: Normal X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/2 X-Sequence-Number: 2 subscribe From pgsql-performance-owner@postgresql.org Tue Sep 10 11:20:22 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 8FE8D47671C for ; Tue, 10 Sep 2002 11:20:19 -0400 (EDT) Received: from mail.arcamax.com (mail.arcamax.com [209.96.210.69]) by postgresql.org (Postfix) with SMTP id 02ABE476711 for ; Tue, 10 Sep 2002 11:20:18 -0400 (EDT) Received: (qmail 11924 invoked by uid 526); 10 Sep 2002 15:20:20 -0000 Received: from unknown (HELO arcamax.com) (192.168.0.32) by 0 with SMTP; 10 Sep 2002 15:20:20 -0000 Message-ID: <3D7E0DB7.9040108@arcamax.com> Date: Tue, 10 Sep 2002 11:20:23 -0400 From: Bryan White User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.1) Gecko/20020826 X-Accept-Language: en-us, en MIME-Version: 1.0 To: pgsql-performance@postgresql.org Subject: subscribe Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS perl-11 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/4 X-Sequence-Number: 4 From pgsql-performance-owner@postgresql.org Tue Sep 10 12:02:04 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id B3FB6476534 for ; Tue, 10 Sep 2002 12:02:00 -0400 (EDT) Received: from h009.c021.snv.cp.net (h009.c021.snv.cp.net [209.228.35.179]) by postgresql.org (Postfix) with ESMTP id 2E6BC476515 for ; Tue, 10 Sep 2002 12:02:00 -0400 (EDT) Received: from vengabox.candlefire.org (12.255.22.43) by h009.c021.snv.cp.net (5.6.0.25) (authenticated as jlarson@candlefire.org) id 3D1DA655000EB024 for pgsql-performance@postgresql.org; Tue, 10 Sep 2002 09:02:05 -0700 Date: Tue, 10 Sep 2002 10:07:06 -0600 From: Jason k Larson X-Mailer: The Bat! (v1.53d) Reply-To: Jason k Larson Organization: CandleFire Productions X-Priority: 3 (Normal) Message-ID: <1472247656.20020910100706@candlefire.org> To: pgsql-performance@postgresql.org Subject: subscribe MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/5 X-Sequence-Number: 5 subscribe From pgsql-performance-owner@postgresql.org Tue Sep 10 09:53:58 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 801B74763D4 for ; Tue, 10 Sep 2002 09:53:57 -0400 (EDT) Received: from pop.e-it.com (unknown [216.187.113.82]) by postgresql.org (Postfix) with SMTP id 767EA4762D0 for ; Tue, 10 Sep 2002 09:53:53 -0400 (EDT) Received: (qmail 4651 invoked from network); 10 Sep 2002 13:53:56 -0000 Received: from unknown (HELO did-it.com) (66.246.13.35) by 0 with SMTP; 10 Sep 2002 13:53:56 -0000 Message-ID: <3D7E2526.7040307@did-it.com> Date: Tue, 10 Sep 2002 10:00:22 -0700 From: Ericson Smith User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.1) Gecko/20020826 X-Accept-Language: en-us, en MIME-Version: 1.0 To: pgsql-performance@postgresql.org Subject: subscribe Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/3 X-Sequence-Number: 3 subscribe From pgsql-performance-owner@postgresql.org Tue Sep 10 22:07:14 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id BC01D476503 for ; Tue, 10 Sep 2002 22:07:12 -0400 (EDT) Received: from gateway.heart.com.au (unknown [202.44.184.121]) by postgresql.org (Postfix) with ESMTP id CFF27476314 for ; Tue, 10 Sep 2002 22:07:08 -0400 (EDT) Received: from tardis.heart.com.au (IDENT:root@tardis.heart.com.au [192.168.0.42]) by gateway.heart.com.au (8.11.6/8.11.2) with ESMTP id g8B2ZKI18868 for ; Wed, 11 Sep 2002 12:35:20 +1000 Received: from heart.com.au ([192.168.0.53]) by tardis.heart.com.au (8.9.3/8.8.7) with ESMTP id MAA17630 for ; Wed, 11 Sep 2002 12:07:01 +1000 Message-ID: <3D7EA4E4.40106@heart.com.au> Date: Wed, 11 Sep 2002 12:05:24 +1000 From: Keith Gray User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:0.9.4.1) Gecko/20020508 Netscape6/6.2.3 X-Accept-Language: en-us MIME-Version: 1.0 To: pgsql-performance@postgresql.org Subject: subscribe Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/6 X-Sequence-Number: 6 -- Keith Gray Technical Services Manager Heart Consulting Services From pgsql-performance-owner@postgresql.org Wed Sep 11 04:47:03 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id AFD06475D57 for ; Wed, 11 Sep 2002 04:46:59 -0400 (EDT) Received: from mail.berusoft.li (ns2.berusoft.li [194.208.67.154]) by postgresql.org (Postfix) with SMTP id E3104475E13 for ; Wed, 11 Sep 2002 04:46:55 -0400 (EDT) Received: (qmail 2444 invoked by uid 99); 11 Sep 2002 10:49:33 -0000 Date: 11 Sep 2002 10:49:33 -0000 Message-ID: <20020911104933.2443.qmail@mail.berusoft.li> Reply-To: "BeruSoft AG" From: "BeruSoft AG" To: pgsql-performance@postgresql.org Subject: subscribe X-Mailer: [web.office] by BeruSoft AG MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/7 X-Sequence-Number: 7 From pgsql-performance-owner@postgresql.org Wed Sep 11 04:57:28 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id E5D86475D57 for ; Wed, 11 Sep 2002 04:57:27 -0400 (EDT) Received: from mail.berusoft.li (ns2.berusoft.li [194.208.67.154]) by postgresql.org (Postfix) with SMTP id 9C152475CB6 for ; Wed, 11 Sep 2002 04:57:25 -0400 (EDT) Received: (qmail 2596 invoked by uid 99); 11 Sep 2002 11:00:03 -0000 Date: 11 Sep 2002 11:00:03 -0000 Message-ID: <20020911110003.2595.qmail@mail.berusoft.li> Reply-To: nici.bertschler@berusoft.com From: nici.bertschler@berusoft.com To: pgsql-performance@postgresql.org Subject: subscribe X-Mailer: [web.office] by BeruSoft AG MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/8 X-Sequence-Number: 8 subscribe From pgsql-performance-owner@postgresql.org Wed Sep 11 07:09:17 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 62E4A4759F5 for ; Wed, 11 Sep 2002 07:09:16 -0400 (EDT) Received: from out-mta3.plasa.com (out-mta2.plasa.com [202.134.0.198]) by postgresql.org (Postfix) with ESMTP id B79CA476506 for ; Wed, 11 Sep 2002 07:09:13 -0400 (EDT) Received: from [61.5.9.21] (helo=portab) by out-mta3.plasa.com with smtp (Exim 4.04) id 17p5Mv-000nOk-00 for pgsql-performance@postgresql.org; Wed, 11 Sep 2002 18:09:13 +0700 Message-ID: <006e01c25984$0e0292c0$1509053d@portab> From: "kopra" To: Subject: subscribe Date: Wed, 11 Sep 2002 18:05:35 +0700 MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 6.00.2600.0000 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2600.0000 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/9 X-Sequence-Number: 9 subscribe ------------------------------------------------------------------------------ > Khusus Pelanggan Telepon DIVRE 2, Tekan 166 untuk mendengarkan pesan Anda ------------------------------------------------------------------------------ From pgsql-performance-owner@postgresql.org Fri Sep 13 20:16:03 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id D4C364760BC for ; Fri, 13 Sep 2002 20:16:00 -0400 (EDT) Received: from smtp3.ihug.com.au (smtp3.ihug.com.au [203.109.250.76]) by postgresql.org (Postfix) with ESMTP id EBF80475E22 for ; Fri, 13 Sep 2002 20:15:58 -0400 (EDT) Received: from p568-tnt1.mel.ihug.com.au (postgresql.org) [203.173.162.60] by smtp3.ihug.com.au with esmtp (Exim 3.22 #1 (Debian)) id 17q0bQ-00077B-00; Sat, 14 Sep 2002 10:16:00 +1000 Message-ID: <3D827FBF.4F137D2E@postgresql.org> Date: Sat, 14 Sep 2002 10:15:59 +1000 From: Justin Clift X-Mailer: Mozilla 4.79 [en] (WinNT; U) X-Accept-Language: en MIME-Version: 1.0 To: PostgreSQL Performance Mailing List Subject: Anyone have any find grained benchmark data? Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/10 X-Sequence-Number: 10 Hi everyone, There are PostgreSQL servers around that are handling 2,000 simultaneous client connections (in real life) without problems, but no-one obvious seems to have yet taken the time to do fine grained testing of the servers which can take this kind of load, to accurately model their performance characteristics. Does anyone here happen to have fine grained benchmark/performance figures hanging around which get into this range of performance? Preferably with pretty precise details of how the system was configured, etc. :-) Regards and best wishes, Justin Clift -- "My grandfather once told me that there are two kinds of people: those who work and those who take the credit. He told me to try to be in the first group; there was less competition there." - Indira Gandhi From pgsql-performance-owner@postgresql.org Tue Sep 17 06:50:17 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 9B7E1475B33 for ; Tue, 17 Sep 2002 06:50:15 -0400 (EDT) Received: from out-mta3.plasa.com (out-mta3.plasa.com [202.134.0.198]) by postgresql.org (Postfix) with ESMTP id CC016475D91 for ; Tue, 17 Sep 2002 06:50:13 -0400 (EDT) Received: from [61.5.16.47] (helo=portab) by out-mta3.plasa.com with smtp (Exim 4.10) id 17rFvo-000YjM-00 for pgsql-performance@postgresql.org; Tue, 17 Sep 2002 17:50:12 +0700 Message-ID: <000101c25e38$66df87a0$2f10053d@portab> From: "kopra" To: Subject: subscribe Date: Sat, 14 Sep 2002 19:28:03 +0700 MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 6.00.2600.0000 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2600.0000 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/11 X-Sequence-Number: 11 subscribe ------------------------------------------------------------------------------ > Khusus Pelanggan Telepon DIVRE 2, Tekan 166 untuk mendengarkan pesan Anda ------------------------------------------------------------------------------ From pgsql-general-owner@postgresql.org Thu Sep 26 04:35:33 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 51FFF4763DE for ; Thu, 26 Sep 2002 04:35:20 -0400 (EDT) Received: from www.pspl.co.in (www.pspl.co.in [202.54.11.65]) by postgresql.org (Postfix) with ESMTP id 9222E476065 for ; Thu, 26 Sep 2002 04:35:15 -0400 (EDT) Received: (from root@localhost) by www.pspl.co.in (8.11.6/8.11.6) id g8Q8aVq12622 for ; Thu, 26 Sep 2002 14:06:31 +0530 Received: from daithan (daithan.intranet.pspl.co.in [192.168.7.161]) by www.pspl.co.in (8.11.6/8.11.0) with ESMTP id g8Q8aUv12612; Thu, 26 Sep 2002 14:06:31 +0530 From: "Shridhar Daithankar" To: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Date: Thu, 26 Sep 2002 14:05:44 +0530 MIME-Version: 1.0 Subject: Performance while loading data and indexing Reply-To: shridhar_daithankar@persistent.co.in Message-ID: <3D931438.22010.133ADAFA@localhost> X-mailer: Pegasus Mail for Windows (v4.02) Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT Content-description: Mail message body X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1161 X-Sequence-Number: 30742 Hello all, Some time back I posted a query to build a site with 150GB of database. In last couple of weeks, lots of things were tested at my place and there are some results and again some concerns. This is a long post. Please be patient and read thr. If we win this, I guess we have a good marketing/advocacy case here..;-) First the problems (For those who do not read beyond first page) 1) Database load time from flat file using copy is very high 2) Creating index takes huge amount of time. 3) Any suggsestions for runtime as data load and query will be going in parallel. Now the details. Note that this is a test run only.. Platform:- 4x Xeon2.4GHz/4GB RAM/4x48 SCSI RAID5/72 GB SCSI RedHat7.2/PostgreSQL7.1.3 Database in flat file: 125,000,000 records of around 100 bytes each. Flat file size 12GB Load time: 14581 sec/~8600 rows persec/~ an MB of data per sec. Create unique composite index on 2 char and a timestamp field: 25226 sec. Database size on disk: 26GB Select query: 1.5 sec. for approx. 150 rows. Important postgresql.conf settings sort_mem = 12000 shared_buffers = 24000 fsync=true (Sad but true. Left untouched.. Will that make a difference on SCSI?) wal_buffers = 65536 wal_files = 64 Now the requirements Initial flat data load: 250GB of data. This has gone up since last query. It was 150GB earlier.. Ongoing inserts: 5000/sec. Number of queries: 4800 queries/hour Query response time: 10 sec. Now questions. 1) Instead of copying from a single 12GB data file, will a parallel copy from say 5 files will speed up the things? Couple MB of data per sec. to disk is just not saturating it. It's a RAID 5 setup.. 2) Sort mem.=12K i.e. 94MB, sounds good enough to me. Does this need further addition to improve create index performance? 3) 5K concurrent inserts with an index on, will this need a additional CPU power? Like deploying it on dual RISC CPUs etc? 4) Query performance is not a problem. Though 4.8K queries per sec. expected response time from each query is 10 sec. But my guess is some serius CPU power will be chewed there too.. 5)Will upgrading to 7.2.2/7.3 beta help? All in all, in the test, we didn't see the performance where hardware is saturated to it's limits. So effectively we are not able to get postgresql making use of it. Just pushing WAL and shared buffers does not seem to be the solution. If you guys have any suggestions. let me know. I need them all.. Mysql is almost out because it's creating index for last 17 hours. I don't think it will keep up with 5K inserts per sec. with index. SAP DB is under evaluation too. But postgresql is most favourite as of now because it works. So I need to come up with solutions to problems that will occur in near future.. ;-) TIA.. Bye Shridhar -- Law of Procrastination: Procrastination avoids boredom; one never has the feeling that there is nothing important to do. From pgsql-general-owner@postgresql.org Thu Sep 26 04:53:43 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id E56844768FD for ; Thu, 26 Sep 2002 04:53:40 -0400 (EDT) Received: from www.pspl.co.in (www.pspl.co.in [202.54.11.65]) by postgresql.org (Postfix) with ESMTP id 2B51C4767DC for ; Thu, 26 Sep 2002 04:53:39 -0400 (EDT) Received: (from root@localhost) by www.pspl.co.in (8.11.6/8.11.6) id g8Q8snQ14266 for ; Thu, 26 Sep 2002 14:24:49 +0530 Received: from daithan (daithan.intranet.pspl.co.in [192.168.7.161]) by www.pspl.co.in (8.11.6/8.11.0) with ESMTP id g8Q8snv14256; Thu, 26 Sep 2002 14:24:49 +0530 From: "Shridhar Daithankar" To: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Date: Thu, 26 Sep 2002 14:24:02 +0530 MIME-Version: 1.0 Subject: Re: [HACKERS] Performance while loading data and indexing Reply-To: shridhar_daithankar@persistent.co.in Message-ID: <3D931882.31859.134B9E4C@localhost> In-reply-to: <3D931438.22010.133ADAFA@localhost> X-mailer: Pegasus Mail for Windows (v4.02) Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT Content-description: Mail message body X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1162 X-Sequence-Number: 30743 On 26 Sep 2002 at 14:05, Shridhar Daithankar wrote: > Some time back I posted a query to build a site with 150GB of database. In last > couple of weeks, lots of things were tested at my place and there are some > results and again some concerns. > 2) Creating index takes huge amount of time. > Load time: 14581 sec/~8600 rows persec/~ an MB of data per sec. > Create unique composite index on 2 char and a timestamp field: 25226 sec. > Database size on disk: 26GB > Select query: 1.5 sec. for approx. 150 rows. > 2) Sort mem.=12K i.e. 94MB, sounds good enough to me. Does this need further > addition to improve create index performance? Just a thought. If I sort the table before making an index, would it be faster than creating index on raw table? And/or if at all, how do I sort the table without duplicating it? Just a wild thought.. Bye Shridhar -- linux: the choice of a GNU generation(ksh@cis.ufl.edu put this on Tshirts in '93) From pgsql-general-owner@postgresql.org Thu Sep 26 05:06:18 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id A48884761EA; Thu, 26 Sep 2002 05:06:16 -0400 (EDT) Received: from svana.org (t1-1-076.dialup.apex.net.au [203.20.62.76]) by postgresql.org (Postfix) with ESMTP id B99B24760CD; Thu, 26 Sep 2002 05:06:15 -0400 (EDT) Received: from kleptog by svana.org with local (Exim 3.35 #1 (Debian)) id 17uUaF-0002nB-00; Thu, 26 Sep 2002 19:05:19 +1000 Date: Thu, 26 Sep 2002 19:05:19 +1000 From: Martijn van Oosterhout To: Shridhar Daithankar Cc: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: Performance while loading data and indexing Message-ID: <20020926090519.GB10471@svana.org> Reply-To: Martijn van Oosterhout Mail-Followup-To: Shridhar Daithankar , pgsql-hackers@postgresql.org, pgsql-general@postgresql.org References: <3D931438.22010.133ADAFA@localhost> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <3D931438.22010.133ADAFA@localhost> User-Agent: Mutt/1.3.28i X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1163 X-Sequence-Number: 30744 I'll preface this by saying that while I have a large database, it doesn't require quite the performace you're talking about here. On Thu, Sep 26, 2002 at 02:05:44PM +0530, Shridhar Daithankar wrote: > 1) Database load time from flat file using copy is very high > 2) Creating index takes huge amount of time. > 3) Any suggsestions for runtime as data load and query will be going in > parallel. You're loading all the data in one copy. I find that INSERTs are mostly limited by indexes. While index lookups are cheap, they are not free and each index needs to be updated for each row. I fond using partial indexes to only index the rows you actually use can help with the loading. It's a bit obscure though. As for parallel loading, you'll be limited mostly by your I/O bandwidth. Have you measured it to take sure it's up to speed? > Now the details. Note that this is a test run only.. > > Platform:- 4x Xeon2.4GHz/4GB RAM/4x48 SCSI RAID5/72 GB SCSI > RedHat7.2/PostgreSQL7.1.3 > > Database in flat file: > 125,000,000 records of around 100 bytes each. > Flat file size 12GB > > Load time: 14581 sec/~8600 rows persec/~ an MB of data per sec. > Create unique composite index on 2 char and a timestamp field: 25226 sec. > Database size on disk: 26GB > Select query: 1.5 sec. for approx. 150 rows. So you're loading at a rate of 860KB per sec. That's not too fast. How many indexes are active at that time? Triggers and foreign keys also take their toll. > Important postgresql.conf settings > > sort_mem = 12000 > shared_buffers = 24000 > fsync=true (Sad but true. Left untouched.. Will that make a difference on > SCSI?) > wal_buffers = 65536 > wal_files = 64 fsync IIRC only affects the WAL buffers now but it may be quite expensive, especially considering it's running on every transaction commit. Oh, your WAL files are on a seperate disk from the data? > Initial flat data load: 250GB of data. This has gone up since last query. It > was 150GB earlier.. > Ongoing inserts: 5000/sec. > Number of queries: 4800 queries/hour > Query response time: 10 sec. That looks quite acheivable. > 1) Instead of copying from a single 12GB data file, will a parallel copy from > say 5 files will speed up the things? Limited by I/O bandwidth. On linux vmstat can tell you how many blocks are being loaded and stored per second. Try it. As long as sync() doesn't get done too often, it should be help. > Couple MB of data per sec. to disk is just not saturating it. It's a RAID 5 > setup.. No, it's not. You should be able to do better. > 2) Sort mem.=12K i.e. 94MB, sounds good enough to me. Does this need further > addition to improve create index performance? Should be fine. Admittedly your indexes are taking rather long to build. > 3) 5K concurrent inserts with an index on, will this need a additional CPU > power? Like deploying it on dual RISC CPUs etc? It shouldn't. Do you have an idea of what your CPU usage is? ps aux should give you a decent idea. > 4) Query performance is not a problem. Though 4.8K queries per sec. expected > response time from each query is 10 sec. But my guess is some serius CPU power > will be chewed there too.. Should be fine. > 5)Will upgrading to 7.2.2/7.3 beta help? Possibly, though it may be wirth it just for the features/bugfixes. > All in all, in the test, we didn't see the performance where hardware is > saturated to it's limits. So effectively we are not able to get postgresql > making use of it. Just pushing WAL and shared buffers does not seem to be the > solution. > > If you guys have any suggestions. let me know. I need them all.. Find the bottleneck: CPU, I/O or memory? > Mysql is almost out because it's creating index for last 17 hours. I don't > think it will keep up with 5K inserts per sec. with index. SAP DB is under > evaluation too. But postgresql is most favourite as of now because it works. So > I need to come up with solutions to problems that will occur in near future.. > ;-) 17 hours! Ouch. Either way, you should be able to do much better. Hope this helps, -- Martijn van Oosterhout http://svana.org/kleptog/ > There are 10 kinds of people in the world, those that can do binary > arithmetic and those that can't. From pgsql-hackers-owner@postgresql.org Thu Sep 26 05:12:58 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 34B1A47616F for ; Thu, 26 Sep 2002 05:12:56 -0400 (EDT) Received: from www.pspl.co.in (www.pspl.co.in [202.54.11.65]) by postgresql.org (Postfix) with ESMTP id 3DF80475F47 for ; Thu, 26 Sep 2002 05:12:53 -0400 (EDT) Received: (from root@localhost) by www.pspl.co.in (8.11.6/8.11.6) id g8Q9E6m15849 for ; Thu, 26 Sep 2002 14:44:06 +0530 Received: from daithan (daithan.intranet.pspl.co.in [192.168.7.161]) by www.pspl.co.in (8.11.6/8.11.0) with ESMTP id g8Q9E6v15839; Thu, 26 Sep 2002 14:44:06 +0530 From: "Shridhar Daithankar" To: paolo.cassago@talentmanager.com Date: Thu, 26 Sep 2002 14:43:20 +0530 MIME-Version: 1.0 Subject: Re: Performance while loading data and indexing Reply-To: shridhar_daithankar@persistent.co.in Cc: pgsql-hackers@postgresql.org Message-ID: <3D931D08.1695.135D474B@localhost> In-reply-to: <19138.194.185.48.247.1033030286.squirrel@mail.talentwebsolutions.com> References: <3D931882.31859.134B9E4C@localhost> X-mailer: Pegasus Mail for Windows (v4.02) Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT Content-description: Mail message body X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1525 X-Sequence-Number: 29482 On 26 Sep 2002 at 10:51, paolo.cassago@talentmanager.c wrote: > Hi, > it seems you have to cluster it, I don't think you have another choise. Hmm.. That didn't occur to me...I guess some real time clustering like usogres would do. Unless it turns out to be a performance hog.. But this is just insert and select. No updates no deletes(Unless customer makes a 180 degree turn) So I doubt if clustering will help. At the most I can replicate data across machines and spread queries on them. Replication overhead as a down side and low query load on each machine as upside.. > I'm retrieving the configuration of our postgres servers (I'm out of office > now), so I can send it to you. I was quite disperate about performance, and > I was thinking to migrate the data on an oracle database. Then I found this > configuration on the net, and I had a succesfull increase of performance. In this case, we are upto postgresql because we/our customer wants to keep the costs down..:-) Even they are asking now if it's possible to keep hardware costs down as well. That's getting some funny responses here but I digress.. > Maybe this can help you. > > Why you use copy to insert records? I usually use perl scripts, and they > work well . Performance reasons. As I said in one of my posts earlier, putting upto 100K records in one transaction in steps of 10K did not reach performance of copy. As Tom said rightly, it was a 4-1 ratio despite using transactions.. Thanks once again.. Bye Shridhar -- Secretary's Revenge: Filing almost everything under "the". From pgsql-general-owner@postgresql.org Thu Sep 26 05:17:41 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id C6DFF4763DE; Thu, 26 Sep 2002 05:17:39 -0400 (EDT) Received: from new-smtp2.ihug.com.au (new-smtp2.ihug.com.au [203.109.250.28]) by postgresql.org (Postfix) with ESMTP id 24E4847631F; Thu, 26 Sep 2002 05:17:39 -0400 (EDT) Received: from p555-tnt2.mel.ihug.com.au (postgresql.org) [203.173.166.47] by new-smtp2.ihug.com.au with esmtp (Exim 3.22 #1 (Debian)) id 17uUm7-00005d-00; Thu, 26 Sep 2002 19:17:35 +1000 Message-ID: <3D92D0AC.CE6114C4@postgresql.org> Date: Thu, 26 Sep 2002 19:17:32 +1000 From: Justin Clift X-Mailer: Mozilla 4.79 [en] (WinNT; U) X-Accept-Language: en MIME-Version: 1.0 To: shridhar_daithankar@persistent.co.in Cc: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: Performance while loading data and indexing References: <3D931438.22010.133ADAFA@localhost> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1164 X-Sequence-Number: 30745 Hi Shridhar, Shridhar Daithankar wrote: > 3) Any suggsestions for runtime as data load and query will be going in > parallel. That sounds unusual. From reading this, it *sounds* like you'll be running queries against an incomplete dataset, or maybe just running the queries that affect the tables loaded thus far (during the initial load). > fsync=true (Sad but true. Left untouched.. Will that make a difference on > SCSI?) Definitely. Have directly measured a ~ 2x tps throughput increase on FreeBSD when leaving fsync off whilst performance measuring stuff recently (PG 7.2.2). Like anything it'll depend on workload, phase of moon, etc, but it's a decent indicator. > Now questions. > > 1) Instead of copying from a single 12GB data file, will a parallel copy from > say 5 files will speed up the things? Not sure yet. Haven't get done enough performance testing (on the cards very soon though). > Couple MB of data per sec. to disk is just not saturating it. It's a RAID 5 > setup.. fsync = off would help during the data load, but not a good idea if you're going to be running queries against it at the same time. Am still getting the hang of performance tuning stuff. Have a bunch of Ultra160 hardware for the Intel platform, and am testing against it as time permits. Not as high end as I'd like, but it's a start. :-) Regards and best wishes, Justin Clift > Bye > Shridhar -- "My grandfather once told me that there are two kinds of people: those who work and those who take the credit. He told me to try to be in the first group; there was less competition there." - Indira Gandhi From pgsql-general-owner@postgresql.org Thu Sep 26 05:35:14 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id C25DF47644C for ; Thu, 26 Sep 2002 05:35:13 -0400 (EDT) Received: from www.pspl.co.in (www.pspl.co.in [202.54.11.65]) by postgresql.org (Postfix) with ESMTP id 0925D47616F for ; Thu, 26 Sep 2002 05:35:10 -0400 (EDT) Received: (from root@localhost) by www.pspl.co.in (8.11.6/8.11.6) id g8Q9aQg18125 for ; Thu, 26 Sep 2002 15:06:26 +0530 Received: from daithan (daithan.intranet.pspl.co.in [192.168.7.161]) by www.pspl.co.in (8.11.6/8.11.0) with ESMTP id g8Q9aQv18115; Thu, 26 Sep 2002 15:06:26 +0530 From: "Shridhar Daithankar" To: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Date: Thu, 26 Sep 2002 15:05:40 +0530 MIME-Version: 1.0 Subject: Re: Performance while loading data and indexing Reply-To: shridhar_daithankar@persistent.co.in Message-ID: <3D932244.13502.1371B9CA@localhost> In-reply-to: <3D92D0AC.CE6114C4@postgresql.org> X-mailer: Pegasus Mail for Windows (v4.02) Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT Content-description: Mail message body X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1166 X-Sequence-Number: 30747 On 26 Sep 2002 at 19:17, Justin Clift wrote: > Shridhar Daithankar wrote: > > > 3) Any suggsestions for runtime as data load and query will be going in > > parallel. > > That sounds unusual. From reading this, it *sounds* like you'll be > running queries against an incomplete dataset, or maybe just running the > queries that affect the tables loaded thus far (during the initial > load). That's correct. Load the data so far and keep inserting data as and when it generates. They don't mind running against data so far. It's not very accurate stuff IMO... > > fsync=true (Sad but true. Left untouched.. Will that make a difference on > > SCSI?) > > Definitely. Have directly measured a ~ 2x tps throughput increase on > FreeBSD when leaving fsync off whilst performance measuring stuff > recently (PG 7.2.2). Like anything it'll depend on workload, phase of > moon, etc, but it's a decent indicator. I didn't know even that matters with SCSI..Will check out.. > fsync = off would help during the data load, but not a good idea if > you're going to be running queries against it at the same time. That's OK for the reasons mentioned above. It wouldn't be out of place to expect a UPS to such an installation... Bye Shridhar -- Hoare's Law of Large Problems: Inside every large problem is a small problem struggling to get out. From pgsql-general-owner@postgresql.org Thu Sep 26 05:46:29 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 75E39476B20 for ; Thu, 26 Sep 2002 05:46:27 -0400 (EDT) Received: from www.pspl.co.in (www.pspl.co.in [202.54.11.65]) by postgresql.org (Postfix) with ESMTP id 548F847644C for ; Thu, 26 Sep 2002 05:46:23 -0400 (EDT) Received: (from root@localhost) by www.pspl.co.in (8.11.6/8.11.6) id g8Q9la019303 for ; Thu, 26 Sep 2002 15:17:36 +0530 Received: from daithan (daithan.intranet.pspl.co.in [192.168.7.161]) by www.pspl.co.in (8.11.6/8.11.0) with ESMTP id g8Q9lav19293; Thu, 26 Sep 2002 15:17:36 +0530 From: "Shridhar Daithankar" To: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Date: Thu, 26 Sep 2002 15:16:50 +0530 MIME-Version: 1.0 Subject: Re: Performance while loading data and indexing Reply-To: shridhar_daithankar@persistent.co.in Message-ID: <3D9324E2.30195.137BF348@localhost> In-reply-to: <20020926090519.GB10471@svana.org> References: <3D931438.22010.133ADAFA@localhost> X-mailer: Pegasus Mail for Windows (v4.02) Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT Content-description: Mail message body X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1167 X-Sequence-Number: 30748 On 26 Sep 2002 at 19:05, Martijn van Oosterhout wrote: > On Thu, Sep 26, 2002 at 02:05:44PM +0530, Shridhar Daithankar wrote: > > 1) Database load time from flat file using copy is very high > > 2) Creating index takes huge amount of time. > > 3) Any suggsestions for runtime as data load and query will be going in > > parallel. > > You're loading all the data in one copy. I find that INSERTs are mostly > limited by indexes. While index lookups are cheap, they are not free and > each index needs to be updated for each row. > > I fond using partial indexes to only index the rows you actually use can > help with the loading. It's a bit obscure though. > > As for parallel loading, you'll be limited mostly by your I/O bandwidth. > Have you measured it to take sure it's up to speed? Well. It's like this, as of now.. CreateDB->create table->create index->Select. So loading is not slowed by index. As of your hint of vmstat, will check it out. > So you're loading at a rate of 860KB per sec. That's not too fast. How many > indexes are active at that time? Triggers and foreign keys also take their > toll. Nothing except the table where data os loaded.. > fsync IIRC only affects the WAL buffers now but it may be quite expensive, > especially considering it's running on every transaction commit. Oh, your > WAL files are on a seperate disk from the data? No. Same RAID 5 disks.. > It shouldn't. Do you have an idea of what your CPU usage is? ps aux should > give you a decent idea. I guess we forgot to monitor system parameters. Next on my list is running vmstat, top and tuning bdflush. > Find the bottleneck: CPU, I/O or memory? Understood.. > > > Mysql is almost out because it's creating index for last 17 hours. I don't > > think it will keep up with 5K inserts per sec. with index. SAP DB is under > > evaluation too. But postgresql is most favourite as of now because it works. So > > I need to come up with solutions to problems that will occur in near future.. > > ;-) > > 17 hours! Ouch. Either way, you should be able to do much better. Hope this > helps, Heh.. no wonder this evaluation is taking more than 2 weeks.. Mysql was running out of disk space while creating index and crashin. An upgrade to mysql helped there but no numbers as yet.. Thanks once again... Bye Shridhar -- Boren's Laws: (1) When in charge, ponder. (2) When in trouble, delegate. (3) When in doubt, mumble. From pgsql-general-owner@postgresql.org Thu Sep 26 05:59:51 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 13499476C17 for ; Thu, 26 Sep 2002 05:59:51 -0400 (EDT) Received: from anchor-post-39.mail.demon.net (anchor-post-39.mail.demon.net [194.217.242.80]) by postgresql.org (Postfix) with ESMTP id 76F3C476C01 for ; Thu, 26 Sep 2002 05:59:50 -0400 (EDT) Received: from mwynhau.demon.co.uk ([193.237.186.96] helo=mainbox.archonet.com) by anchor-post-39.mail.demon.net with esmtp (Exim 3.36 #2) id 17uVR1-0003WF-0U; Thu, 26 Sep 2002 10:59:51 +0100 Received: from localhost (localhost.localdomain [127.0.0.1]) by mainbox.archonet.com (Postfix) with ESMTP id 5B12716378; Thu, 26 Sep 2002 10:49:02 +0100 (BST) Received: from client.archonet.com (client.archonet.com [192.168.1.16]) by mainbox.archonet.com (Postfix) with ESMTP id AE86216367; Thu, 26 Sep 2002 10:49:01 +0100 (BST) Content-Type: text/plain; charset="iso-8859-1" From: Richard Huxton Organization: Archonet Ltd To: shridhar_daithankar@persistent.co.in, pgsql-general@postgresql.org Subject: Re: Performance while loading data and indexing Date: Thu, 26 Sep 2002 10:48:06 +0100 User-Agent: KMail/1.4.1 References: <3D931438.22010.133ADAFA@localhost> In-Reply-To: <3D931438.22010.133ADAFA@localhost> MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Message-Id: <200209261048.07761.dev@archonet.com> X-Virus-Scanned: by AMaViS snapshot-20020531 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1172 X-Sequence-Number: 30753 On Thursday 26 Sep 2002 9:35 am, Shridhar Daithankar wrote: [questions re: large database] Before reading my advice please bear in mind you are operating way beyond t= he=20 scale of anything I have ever built. > Now the details. Note that this is a test run only.. > > Platform:- 4x Xeon2.4GHz/4GB RAM/4x48 SCSI RAID5/72 GB SCSI > RedHat7.2/PostgreSQL7.1.3 > > Database in flat file: > 125,000,000 records of around 100 bytes each. > Flat file size 12GB > > Load time: 14581 sec/~8600 rows persec/~ an MB of data per sec. > Create unique composite index on 2 char and a timestamp field: 25226 sec. > Database size on disk: 26GB > Select query: 1.5 sec. for approx. 150 rows. > > Important postgresql.conf settings [snipped setting details for moment] Have you tried putting the wal files, syslog etc on separate disks/volumes?= If=20 you've settled on Intel, about the only thing you can optimise further is t= he=20 disks. Oh - and the OS - make sure you're running a (good) recent kernel for that= =20 sort of hardware, I seem to remember some substantial changes in the 2.4=20 series regarding multi-processor. > Now the requirements > > Initial flat data load: 250GB of data. This has gone up since last query. > It was 150GB earlier.. > Ongoing inserts: 5000/sec. > Number of queries: 4800 queries/hour > Query response time: 10 sec. Is this 5000 rows in say 500 transactions or 5000 insert transactions per= =20 second. How many concurrent clients is this? Similarly for the 4800 queries= ,=20 how many concurrent clients is this? Are they expected to return approx 150= =20 rows as in your test? > Now questions. > > 1) Instead of copying from a single 12GB data file, will a parallel copy > from say 5 files will speed up the things? If the CPU is the bottle-neck then it should, but it's difficult to say=20 without figures. > Couple MB of data per sec. to disk is just not saturating it. It's a RAID= 5 > setup.. What is saturating during the flat-file load? Something must be maxed in to= p /=20 iostat / vmstat. [snip] > > 5)Will upgrading to 7.2.2/7.3 beta help? It's unlikely to hurt. > All in all, in the test, we didn't see the performance where hardware is > saturated to it's limits. Something *must* be. What are your disaster recovery plans? I can see problems with taking backu= ps=20 if this beast is live 24/7. - Richard Huxton From pgsql-general-owner@postgresql.org Thu Sep 26 05:50:02 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 6D16F4761DF; Thu, 26 Sep 2002 05:50:00 -0400 (EDT) Received: from new-smtp2.ihug.com.au (new-smtp2.ihug.com.au [203.109.250.28]) by postgresql.org (Postfix) with ESMTP id 79EAE47616F; Thu, 26 Sep 2002 05:49:59 -0400 (EDT) Received: from p555-tnt2.mel.ihug.com.au (postgresql.org) [203.173.166.47] by new-smtp2.ihug.com.au with esmtp (Exim 3.22 #1 (Debian)) id 17uVHP-000184-00; Thu, 26 Sep 2002 19:49:55 +1000 Message-ID: <3D92D841.3E02B2A8@postgresql.org> Date: Thu, 26 Sep 2002 19:49:53 +1000 From: Justin Clift X-Mailer: Mozilla 4.79 [en] (WinNT; U) X-Accept-Language: en MIME-Version: 1.0 To: shridhar_daithankar@persistent.co.in Cc: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: [HACKERS] Performance while loading data and indexing References: <3D932244.13502.1371B9CA@localhost> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1168 X-Sequence-Number: 30749 Shridhar Daithankar wrote: > > > fsync=true (Sad but true. Left untouched.. Will that make a difference on > > > SCSI?) > > > > Definitely. Have directly measured a ~ 2x tps throughput increase on > > FreeBSD when leaving fsync off whilst performance measuring stuff > > recently (PG 7.2.2). Like anything it'll depend on workload, phase of > > moon, etc, but it's a decent indicator. > > I didn't know even that matters with SCSI..Will check out.. Cool. When testing it had FreeBSD 4.6.2 installed on one drive along with the PostgreSQL 7.2.2 binaries, it had the data on a second drive (mounted as /pgdata), and it had the pg_xlog directory mounted on a third drive. Swap had it's own drive as well. Everything is UltraSCSI, etc. Haven't yet tested for a performance difference through moving the indexes to another drive after creation though. That apparently has the potential to help as well. :-) Regards and best wishes, Justin Clift -- "My grandfather once told me that there are two kinds of people: those who work and those who take the credit. He told me to try to be in the first group; there was less competition there." - Indira Gandhi From pgsql-general-owner@postgresql.org Thu Sep 26 05:56:37 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 7E7F9476B72; Thu, 26 Sep 2002 05:56:36 -0400 (EDT) Received: from new-smtp2.ihug.com.au (new-smtp2.ihug.com.au [203.109.250.28]) by postgresql.org (Postfix) with ESMTP id 4AB45476B45; Thu, 26 Sep 2002 05:56:34 -0400 (EDT) Received: from p555-tnt2.mel.ihug.com.au (postgresql.org) [203.173.166.47] by new-smtp2.ihug.com.au with esmtp (Exim 3.22 #1 (Debian)) id 17uVNr-0001Ml-00; Thu, 26 Sep 2002 19:56:35 +1000 Message-ID: <3D92D9D2.64CF55F7@postgresql.org> Date: Thu, 26 Sep 2002 19:56:34 +1000 From: Justin Clift X-Mailer: Mozilla 4.79 [en] (WinNT; U) X-Accept-Language: en MIME-Version: 1.0 To: shridhar_daithankar@persistent.co.in Cc: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: [HACKERS] Performance while loading data and indexing References: <3D931438.22010.133ADAFA@localhost> <3D9324E2.30195.137BF348@localhost> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1171 X-Sequence-Number: 30752 Shridhar Daithankar wrote: > > On 26 Sep 2002 at 19:05, Martijn van Oosterhout wrote: > > fsync IIRC only affects the WAL buffers now but it may be quite expensive, > > especially considering it's running on every transaction commit. Oh, your > > WAL files are on a seperate disk from the data? > > No. Same RAID 5 disks.. Not sure if this is a good idea. Would have to think deeply about the controller and drive optimisation/load characteristics. If it's any help, when I was testing recently with WAL on a separate drive, the WAL logs were doing more read&writes per second than the main data drive. This would of course be affected by the queries you are running against the database. I was just running Tatsuo's TPC-B stuff, and the OSDB AS3AP tests. > I guess we forgot to monitor system parameters. Next on my list is running > vmstat, top and tuning bdflush. That'll just be the start of it for serious performance tuning and learning how PostgreSQL works. :) > Thanks once again... > Bye > Shridhar -- "My grandfather once told me that there are two kinds of people: those who work and those who take the credit. He told me to try to be in the first group; there was less competition there." - Indira Gandhi From pgsql-general-owner@postgresql.org Thu Sep 26 10:34:21 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 05F75476AA7; Thu, 26 Sep 2002 10:34:20 -0400 (EDT) Received: from sss.pgh.pa.us (unknown [192.204.191.242]) by postgresql.org (Postfix) with ESMTP id 6AFEC4767D7; Thu, 26 Sep 2002 10:34:18 -0400 (EDT) Received: from sss2.sss.pgh.pa.us (tgl@localhost [127.0.0.1]) by sss.pgh.pa.us (8.12.5/8.12.5) with ESMTP id g8QEXwhR003937; Thu, 26 Sep 2002 10:33:58 -0400 (EDT) To: shridhar_daithankar@persistent.co.in Cc: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: Performance while loading data and indexing In-reply-to: <3D931438.22010.133ADAFA@localhost> References: <3D931438.22010.133ADAFA@localhost> Comments: In-reply-to "Shridhar Daithankar" message dated "Thu, 26 Sep 2002 14:05:44 +0530" Date: Thu, 26 Sep 2002 10:33:58 -0400 Message-ID: <3936.1033050838@sss.pgh.pa.us> From: Tom Lane X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1181 X-Sequence-Number: 30762 "Shridhar Daithankar" writes: > RedHat7.2/PostgreSQL7.1.3 I'd suggest a newer release of Postgres ... 7.1.3 is pretty old ... > Create unique composite index on 2 char and a timestamp field: 25226 sec. What do you mean by "char" exactly? If it's really char(N), how much are you paying in padding space? There are very very few cases where I'd not say to use varchar(N), or text, instead. Also, does it have to be character data? If you could use an integer or float datatype instead the index operations should be faster (though I can't say by how much). Have you thought carefully about the order in which the composite index columns are listed? > sort_mem = 12000 To create an index of this size, you want to push sort_mem as high as it can go without swapping. 12000 sounds fine for the global setting, but in the process that will create the index, try setting sort_mem to some hundreds of megs or even 1Gb. (But be careful: the calculation of space actually used by CREATE INDEX is off quite a bit in pre-7.3 releases :-(. You should probably expect the actual process size to grow to two or three times what you set sort_mem to. Don't let it get so big as to swap.) > wal_buffers = 65536 The above is a complete waste of memory space, which would be better spent on letting the kernel expand its disk cache. There's no reason for wal_buffers to be more than a few dozen. regards, tom lane From pgsql-general-owner@postgresql.org Thu Sep 26 10:42:11 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 45ACE476D3D; Thu, 26 Sep 2002 10:42:11 -0400 (EDT) Received: from sss.pgh.pa.us (unknown [192.204.191.242]) by postgresql.org (Postfix) with ESMTP id B5B5B476D3A; Thu, 26 Sep 2002 10:42:10 -0400 (EDT) Received: from sss2.sss.pgh.pa.us (tgl@localhost [127.0.0.1]) by sss.pgh.pa.us (8.12.5/8.12.5) with ESMTP id g8QEg8hR004032; Thu, 26 Sep 2002 10:42:09 -0400 (EDT) To: Justin Clift Cc: shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: [HACKERS] Performance while loading data and indexing In-reply-to: <3D92D9D2.64CF55F7@postgresql.org> References: <3D931438.22010.133ADAFA@localhost> <3D9324E2.30195.137BF348@localhost> <3D92D9D2.64CF55F7@postgresql.org> Comments: In-reply-to Justin Clift message dated "Thu, 26 Sep 2002 19:56:34 +1000" Date: Thu, 26 Sep 2002 10:42:08 -0400 Message-ID: <4031.1033051328@sss.pgh.pa.us> From: Tom Lane X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1183 X-Sequence-Number: 30764 Justin Clift writes: >> On 26 Sep 2002 at 19:05, Martijn van Oosterhout wrote: >>> fsync IIRC only affects the WAL buffers now but it may be quite expensive, >>> especially considering it's running on every transaction commit. Oh, your >>> WAL files are on a seperate disk from the data? > Not sure if this is a good idea. Would have to think deeply about the > controller and drive optimisation/load characteristics. > If it's any help, when I was testing recently with WAL on a separate > drive, the WAL logs were doing more read&writes per second than the main > data drive. ... but way fewer seeks. For anything involving lots of updating transactions (and certainly 5000 separate insertions per second would qualify; can those be batched??), it should be a win to put WAL on its own spindle, just to get locality of access to the WAL. regards, tom lane From pgsql-general-owner@postgresql.org Thu Sep 26 10:51:34 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id DEDD4475EE4 for ; Thu, 26 Sep 2002 10:51:33 -0400 (EDT) Received: from www.pspl.co.in (www.pspl.co.in [202.54.11.65]) by postgresql.org (Postfix) with ESMTP id 4CAFB475CB4 for ; Thu, 26 Sep 2002 10:51:32 -0400 (EDT) Received: (from root@localhost) by www.pspl.co.in (8.11.6/8.11.6) id g8QEqqK15165 for ; Thu, 26 Sep 2002 20:22:52 +0530 Received: from daithan (daithan.intranet.pspl.co.in [192.168.7.161]) by www.pspl.co.in (8.11.6/8.11.0) with ESMTP id g8QEqqv15154; Thu, 26 Sep 2002 20:22:52 +0530 From: "Shridhar Daithankar" To: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Date: Thu, 26 Sep 2002 20:22:05 +0530 MIME-Version: 1.0 Subject: Re: Performance while loading data and indexing Reply-To: shridhar_daithankar@persistent.co.in Message-ID: <3D936C6D.12380.14936AEC@localhost> References: <3D931438.22010.133ADAFA@localhost> In-reply-to: <3936.1033050838@sss.pgh.pa.us> X-mailer: Pegasus Mail for Windows (v4.02) Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT Content-description: Mail message body X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1187 X-Sequence-Number: 30768 On 26 Sep 2002 at 10:33, Tom Lane wrote: > "Shridhar Daithankar" writes: > > RedHat7.2/PostgreSQL7.1.3 > > I'd suggest a newer release of Postgres ... 7.1.3 is pretty old ... I agree.. downloadind 7.2.2 right away.. > > Create unique composite index on 2 char and a timestamp field: 25226 sec. > > What do you mean by "char" exactly? If it's really char(N), how much > are you paying in padding space? There are very very few cases where > I'd not say to use varchar(N), or text, instead. Also, does it have to > be character data? If you could use an integer or float datatype > instead the index operations should be faster (though I can't say by > how much). Have you thought carefully about the order in which the > composite index columns are listed? I have forwarded the idea of putting things into number. If it causes speedup in index lookup/creation, it would do. Looks like bigint is the order of the day.. > > > sort_mem = 12000 > > To create an index of this size, you want to push sort_mem as high as it > can go without swapping. 12000 sounds fine for the global setting, but > in the process that will create the index, try setting sort_mem to some > hundreds of megs or even 1Gb. (But be careful: the calculation of space > actually used by CREATE INDEX is off quite a bit in pre-7.3 releases > :-(. You should probably expect the actual process size to grow to two > or three times what you set sort_mem to. Don't let it get so big as to > swap.) Great. I was skeptical to push it beyond 100MB. Now I can push it to corners.. > > wal_buffers = 65536 > > The above is a complete waste of memory space, which would be better > spent on letting the kernel expand its disk cache. There's no reason > for wal_buffers to be more than a few dozen. That was a rather desparate move. Nothing was improving performance and then we started pushing numbers.. WIll get it back.. Same goes for 64 WAL files.. A GB looks like waste to me.. I might have found the bottleneck, although by accident. Mysql was running out of space while creating index. So my friend shut down mysql and tried to move things by hand to create links. He noticed that even things like cp were terribly slow and it hit us.. May be the culprit is the file system. Ext3 in this case. My friend argues for ext2 to eliminate journalling overhead but I favour reiserfs personally having used it in pgbench with 10M rows on paltry 20GB IDE disk for 25 tps.. We will be attempting raiserfs and/or XFS if required. I know how much speed difference exists between resiserfs and ext2. Would not be surprised if everythng just starts screaming in one go.. Bye Shridhar -- Cropp's Law: The amount of work done varies inversly with the time spent in the office. From pgsql-general-owner@postgresql.org Thu Sep 26 10:57:42 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 79B71476141 for ; Thu, 26 Sep 2002 10:57:41 -0400 (EDT) Received: from www.pspl.co.in (www.pspl.co.in [202.54.11.65]) by postgresql.org (Postfix) with ESMTP id 0C288475FBD for ; Thu, 26 Sep 2002 10:57:40 -0400 (EDT) Received: (from root@localhost) by www.pspl.co.in (8.11.6/8.11.6) id g8QEww115498 for ; Thu, 26 Sep 2002 20:28:58 +0530 Received: from daithan (daithan.intranet.pspl.co.in [192.168.7.161]) by www.pspl.co.in (8.11.6/8.11.0) with ESMTP id g8QEwwv15488; Thu, 26 Sep 2002 20:28:58 +0530 From: "Shridhar Daithankar" To: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Date: Thu, 26 Sep 2002 20:28:11 +0530 MIME-Version: 1.0 Subject: Re: [HACKERS] Performance while loading data and indexing Reply-To: shridhar_daithankar@persistent.co.in Message-ID: <3D936DDB.26585.14990280@localhost> References: <3D92D9D2.64CF55F7@postgresql.org> In-reply-to: <4031.1033051328@sss.pgh.pa.us> X-mailer: Pegasus Mail for Windows (v4.02) Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT Content-description: Mail message body X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1188 X-Sequence-Number: 30769 On 26 Sep 2002 at 10:42, Tom Lane wrote: > Justin Clift writes: > > If it's any help, when I was testing recently with WAL on a separate > > drive, the WAL logs were doing more read&writes per second than the main > > data drive. > > ... but way fewer seeks. For anything involving lots of updating > transactions (and certainly 5000 separate insertions per second would > qualify; can those be batched??), it should be a win to put WAL on its > own spindle, just to get locality of access to the WAL. Probably they will be a single transcation. If possible we will bunch more of them together.. like 5 seconds of data pushed down in a single transaction but not sure it's possible.. This is bit like replication but from live oracle machine to postgres, from information I have. So there should be some chance of tuning there.. Bye Shridhar -- Langsam's Laws: (1) Everything depends. (2) Nothing is always. (3) Everything is sometimes. From pgsql-general-owner@postgresql.org Thu Sep 26 11:07:07 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id B7B27476D50; Thu, 26 Sep 2002 11:07:06 -0400 (EDT) Received: from mx.webmailstation.com (mx.webmailstation.com [64.23.55.10]) by postgresql.org (Postfix) with ESMTP id 24B2C476D4E; Thu, 26 Sep 2002 11:07:06 -0400 (EDT) Received: from dyp (localhost.localdomain [127.0.0.1]) by mx.webmailstation.com (Postfix) with ESMTP id 202E51F85F; Thu, 26 Sep 2002 07:04:18 -0400 (EDT) Content-Type: text/plain; charset="iso-8859-1" From: Denis Perchine Organization: AcademSoft Ltd. To: shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: [HACKERS] Performance while loading data and indexing Date: Thu, 26 Sep 2002 22:04:41 +0700 User-Agent: KMail/1.4.3 References: <3D931438.22010.133ADAFA@localhost> <3D936C6D.12380.14936AEC@localhost> In-Reply-To: <3D936C6D.12380.14936AEC@localhost> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Message-Id: <200209262204.41638.dyp@perchine.com> X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1189 X-Sequence-Number: 30770 On Thursday 26 September 2002 21:52, Shridhar Daithankar wrote: > I might have found the bottleneck, although by accident. Mysql was running > out of space while creating index. So my friend shut down mysql and tried > to move things by hand to create links. He noticed that even things like cp > were terribly slow and it hit us.. May be the culprit is the file system. > Ext3 in this case. > > My friend argues for ext2 to eliminate journalling overhead but I favour > reiserfs personally having used it in pgbench with 10M rows on paltry 20GB > IDE disk for 25 tps.. > > We will be attempting raiserfs and/or XFS if required. I know how much > speed difference exists between resiserfs and ext2. Would not be surprised > if everythng just starts screaming in one go.. As it was found by someone before any non-journaling FS is faster than journaling one. This due to double work done by FS and database. Try it on ext2 and compare. -- Denis From pgsql-general-owner@postgresql.org Thu Sep 26 11:12:48 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id E29E6476D79; Thu, 26 Sep 2002 11:12:47 -0400 (EDT) Received: from new-smtp2.ihug.com.au (new-smtp2.ihug.com.au [203.109.250.28]) by postgresql.org (Postfix) with ESMTP id 1E9DC476D76; Thu, 26 Sep 2002 11:12:47 -0400 (EDT) Received: from p407-tnt2.mel.ihug.com.au (postgresql.org) [203.173.165.153] by new-smtp2.ihug.com.au with esmtp (Exim 3.22 #1 (Debian)) id 17uaJu-00033K-00; Fri, 27 Sep 2002 01:12:50 +1000 Message-ID: <3D9323F1.3A534EA8@postgresql.org> Date: Fri, 27 Sep 2002 01:12:49 +1000 From: Justin Clift X-Mailer: Mozilla 4.79 [en] (WinNT; U) X-Accept-Language: en MIME-Version: 1.0 To: shridhar_daithankar@persistent.co.in, PostgreSQL Performance Mailing List Cc: PostgreSQL Hackers Mailing List , PostgreSQL General Mailing List Subject: Re: Performance while loading data and indexing References: <3D931438.22010.133ADAFA@localhost> <3D936C6D.12380.14936AEC@localhost> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1191 X-Sequence-Number: 30772 Shridhar Daithankar wrote: > My friend argues for ext2 to eliminate journalling overhead but I favour > reiserfs personally having used it in pgbench with 10M rows on paltry 20GB IDE > disk for 25 tps.. If it's any help, the setup I mentioned before with differnt disks for the data and the WAL files was getting an average of about 72 tps with 200 concurrent users on pgbench. Haven't tuned it in a hard core way at all, and it only has 256MB DDR RAM in it at the moment (single CPU AthonXP 1600). These are figures made during the 2.5k+ test runs of pgbench done when developing pg_autotune recently. As a curiosity point, how predictable are the queries you're going to be running on your database? They sound very simple and very predicatable. The pg_autotune tool might be your friend here. It can deal with arbitrary SQL instead of using the pg_bench stuff of Tatsuos, and it can also deal with an already loaded database. You'd just have to tweak the names of the tables that it vacuums and the names of the indexes that it reindexes between each run, to get some idea of your overall server performance at different load points. Probably worth taking a good look at if you're not afraid of editing variables in C code. :) > We will be attempting raiserfs and/or XFS if required. I know how much speed > difference exists between resiserfs and ext2. Would not be surprised if > everythng just starts screaming in one go.. We'd all probably be interested to hear this. Added the PostgreSQL "Performance" mailing list to this thread too, Just In Case. (wow that's a lot of cross posting now). Regards and best wishes, Justin Clift > Bye > Shridhar > > -- > Cropp's Law: The amount of work done varies inversly with the time spent in the > office. > > ---------------------------(end of broadcast)--------------------------- > TIP 3: if posting/reading through Usenet, please send an appropriate > subscribe-nomail command to majordomo@postgresql.org so that your > message can get through to the mailing list cleanly -- "My grandfather once told me that there are two kinds of people: those who work and those who take the credit. He told me to try to be in the first group; there was less competition there." - Indira Gandhi From pgsql-general-owner@postgresql.org Thu Sep 26 11:28:32 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 48544476CC6 for ; Thu, 26 Sep 2002 11:28:30 -0400 (EDT) Received: from www.pspl.co.in (www.pspl.co.in [202.54.11.65]) by postgresql.org (Postfix) with ESMTP id 42D14476B61 for ; Thu, 26 Sep 2002 11:28:28 -0400 (EDT) Received: (from root@localhost) by www.pspl.co.in (8.11.6/8.11.6) id g8QFTm217943 for ; Thu, 26 Sep 2002 20:59:48 +0530 Received: from daithan (daithan.intranet.pspl.co.in [192.168.7.161]) by www.pspl.co.in (8.11.6/8.11.0) with ESMTP id g8QFTmv17928; Thu, 26 Sep 2002 20:59:48 +0530 From: "Shridhar Daithankar" To: PostgreSQL Performance Mailing List Date: Thu, 26 Sep 2002 20:59:01 +0530 MIME-Version: 1.0 Subject: Re: Performance while loading data and indexing Reply-To: shridhar_daithankar@persistent.co.in Cc: PostgreSQL Hackers Mailing List , PostgreSQL General Mailing List Message-ID: <3D937515.11546.14B53C07@localhost> In-reply-to: <3D9323F1.3A534EA8@postgresql.org> X-mailer: Pegasus Mail for Windows (v4.02) Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT Content-description: Mail message body X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1193 X-Sequence-Number: 30774 On 27 Sep 2002 at 1:12, Justin Clift wrote: > Shridhar Daithankar wrote: > As a curiosity point, how predictable are the queries you're going to be > running on your database? They sound very simple and very predicatable. Mostly predictable selects. Not a domain expert on telecom so not very sure. But in my guess prepare statement in 7.3 should come pretty handy. i.e. by the time we finish evaluation and test deployment, 7.3 will be out in next couple of months to say so. So I would recommend doing it 7.3 way only.. > > The pg_autotune tool might be your friend here. It can deal with > arbitrary SQL instead of using the pg_bench stuff of Tatsuos, and it can > also deal with an already loaded database. You'd just have to tweak the > names of the tables that it vacuums and the names of the indexes that it > reindexes between each run, to get some idea of your overall server > performance at different load points. > > Probably worth taking a good look at if you're not afraid of editing > variables in C code. :) Gladly. We started with altering pgbench here for testing and rapidly settled to perl generated random queries. Once postgresql wins the evaluation match and things come to implementation, pg_autotune would be a handy tool. Just that can't do it right now. Have to fight mysql and SAP DB before that.. BTW any performance figures on SAP DB? People here are as it frustrated with it with difficulties in setting it up. But still.. > > > We will be attempting raiserfs and/or XFS if required. I know how much speed > > difference exists between resiserfs and ext2. Would not be surprised if > > everythng just starts screaming in one go.. > > We'd all probably be interested to hear this. Added the PostgreSQL > "Performance" mailing list to this thread too, Just In Case. (wow that's > a lot of cross posting now). I know..;-) Glad that PG list does not have strict policies like no non- subscriber posting or no attachments.. etc.. IMO reiserfs, though journalling one, is faster than ext2 etc. because the way it handles metadata. Personally I haven't come across ext2 being faster than reiserfs on few machine here for day to day use. I guess I should have a freeBSD CD handy too.. Just to give it a try. If it comes down to a better VM.. though using 2.4.19 here.. so souldn't matter much.. I will keep you guys posted on file system stuff... Glad that we have much flexibility with postgresql.. Bye Shridhar -- Bilbo's First Law: You cannot count friends that are all packed up in barrels. From pgsql-general-owner@postgresql.org Thu Sep 26 11:41:47 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 26ED6476D61; Thu, 26 Sep 2002 11:41:46 -0400 (EDT) Received: from CopelandConsulting.Net (dsl-24293-ld.customer.centurytel.net [209.142.135.135]) by postgresql.org (Postfix) with ESMTP id E26AF476D7B; Thu, 26 Sep 2002 11:41:44 -0400 (EDT) Received: from mouse.copelandconsulting.net (mouse.copelandconsulting.net [192.168.1.2]) by CopelandConsulting.Net (8.10.1/8.10.1) with ESMTP id g8QFfWu10941; Thu, 26 Sep 2002 10:41:32 -0500 (CDT) X-Trade-Id: To: shridhar_daithankar@persistent.co.in Cc: PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List In-Reply-To: <3D936C6D.12380.14936AEC@localhost> References: <3D931438.22010.133ADAFA@localhost> <3D936C6D.12380.14936AEC@localhost> Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="=-u8SGzlKGiTZg+qY+QPln" X-Mailer: Ximian Evolution 1.0.8 Date: 26 Sep 2002 10:41:37 -0500 Message-Id: <1033054898.17282.9.camel@mouse.copelandconsulting.net> Mime-Version: 1.0 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1196 X-Sequence-Number: 30777 --=-u8SGzlKGiTZg+qY+QPln Content-Type: text/plain Content-Transfer-Encoding: quoted-printable On Thu, 2002-09-26 at 09:52, Shridhar Daithankar wrote: > My friend argues for ext2 to eliminate journalling overhead but I favour= =20 > reiserfs personally having used it in pgbench with 10M rows on paltry 20G= B IDE=20 > disk for 25 tps.. >=20 > We will be attempting raiserfs and/or XFS if required. I know how much sp= eed=20 > difference exists between resiserfs and ext2. Would not be surprised if= =20 > everythng just starts screaming in one go.. >=20 I'm not sure about reiserfs or ext3 but with XFS, you can create your log on another disk. Also worth noting is that you can also configure the size and number of log buffers. There are also some other performance type enhancements you can fiddle with if you don't mind risking time stamp consistency in the event of a crash. If your setup allows for it, you might want to consider using XFS in this configuration. While I have not personally tried moving XFS' log to another device, I've heard that performance gains can be truly stellar. Assuming memory allows, twiddling with the log buffering is said to allow for large strides in performance as well. If you do try this, I'd love to hear back about your results and impressions. Greg --=-u8SGzlKGiTZg+qY+QPln Content-Type: application/pgp-signature; name=signature.asc Content-Description: This is a digitally signed message part -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.7 (GNU/Linux) iD8DBQA9kyqx4lr1bpbcL6kRApydAJ46EfAEimKL7eDNSS7ZMdZlo3VptACfcPL0 ByvQOwuqz/14LUVP1Oewjsk= =VHX7 -----END PGP SIGNATURE----- --=-u8SGzlKGiTZg+qY+QPln-- From pgsql-general-owner@postgresql.org Thu Sep 26 12:42:01 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id F1D0F475E83; Thu, 26 Sep 2002 12:41:58 -0400 (EDT) Received: from candle.pha.pa.us (momjian.navpoint.com [207.106.42.251]) by postgresql.org (Postfix) with ESMTP id 3A9EB476EAE; Thu, 26 Sep 2002 12:41:55 -0400 (EDT) Received: (from pgman@localhost) by candle.pha.pa.us (8.11.6/8.10.1) id g8QGfYc04099; Thu, 26 Sep 2002 12:41:34 -0400 (EDT) From: Bruce Momjian Message-Id: <200209261641.g8QGfYc04099@candle.pha.pa.us> Subject: Re: Performance while loading data and indexing In-Reply-To: <3D936C6D.12380.14936AEC@localhost> To: shridhar_daithankar@persistent.co.in Date: Thu, 26 Sep 2002 12:41:34 -0400 (EDT) Cc: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org X-Mailer: ELM [version 2.4ME+ PL99 (25)] MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1202 X-Sequence-Number: 30783 Shridhar Daithankar wrote: > I might have found the bottleneck, although by accident. Mysql was running out > of space while creating index. So my friend shut down mysql and tried to move > things by hand to create links. He noticed that even things like cp were > terribly slow and it hit us.. May be the culprit is the file system. Ext3 in > this case. I just added a file system and multi-cpu section to my performance tuning paper: http://www.ca.postgresql.org/docs/momjian/hw_performance/ The paper does recommend ext3, but the differences between file systems are very small. If you are seeing 'cp' as slow, I wonder if it may be something more general, like poorly tuned hardware or something. You can use 'dd' to throw some data around the file system and see if that is showing slowness; compare those numbers to another machine that has different hardware/OS. Also, though ext3 is slower, turning fsync off should make ext3 function similar to ext2. That would be an interesting test if you suspect ext3. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073 From pgsql-general-owner@postgresql.org Thu Sep 26 13:17:12 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id E040F476E77; Thu, 26 Sep 2002 13:17:08 -0400 (EDT) Received: from varsoon.wireboard.com (www.wireboard.com [216.151.155.101]) by postgresql.org (Postfix) with ESMTP id 8FF09476DFA; Thu, 26 Sep 2002 13:17:07 -0400 (EDT) Received: from doug by varsoon.wireboard.com with local (Exim 3.35 #1) id 17ucFg-0006dh-00; Thu, 26 Sep 2002 13:16:36 -0400 To: Greg Copeland From: Doug cNaught Cc: shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List Subject: Re: [HACKERS] Performance while loading data and indexing References: <3D931438.22010.133ADAFA@localhost> <3D936C6D.12380.14936AEC@localhost> <1033054898.17282.9.camel@mouse.copelandconsulting.net> Date: 26 Sep 2002 13:16:36 -0400 In-Reply-To: Greg Copeland's message of "26 Sep 2002 10:41:37 -0500" Message-ID: Lines: 25 User-Agent: Gnus/5.0806 (Gnus v5.8.6) Emacs/20.7 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1206 X-Sequence-Number: 30787 Greg Copeland writes: > I'm not sure about reiserfs or ext3 but with XFS, you can create your > log on another disk. Also worth noting is that you can also configure > the size and number of log buffers. There are also some other > performance type enhancements you can fiddle with if you don't mind > risking time stamp consistency in the event of a crash. If your setup > allows for it, you might want to consider using XFS in this > configuration. You can definitely put the ext3 log on a different disk with 2.4 kernels. Also, if you put the WAL logs on a different disk from the main database, and mount that partition with 'data=writeback' (ie metadata-only journaling) ext3 should be pretty fast, since WAL files are preallocated and there will therefore be almost no metadata updates. You should be able to mount the main database with "data=ordered" (the default) for good performance and reasonable safety. I think putting WAL on its own disk(s) is one of the keys here. -Doug From pgsql-general-owner@postgresql.org Thu Sep 26 14:14:21 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 8779F476D31; Thu, 26 Sep 2002 13:37:58 -0400 (EDT) Received: from CopelandConsulting.Net (dsl-24293-ld.customer.centurytel.net [209.142.135.135]) by postgresql.org (Postfix) with ESMTP id 61C82476D2E; Thu, 26 Sep 2002 13:37:57 -0400 (EDT) Received: from mouse.copelandconsulting.net (mouse.copelandconsulting.net [192.168.1.2]) by CopelandConsulting.Net (8.10.1/8.10.1) with ESMTP id g8QHapu12099; Thu, 26 Sep 2002 12:36:51 -0500 (CDT) X-Trade-Id: To: Bruce Momjian Cc: shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List In-Reply-To: <200209261641.g8QGfYc04099@candle.pha.pa.us> References: <200209261641.g8QGfYc04099@candle.pha.pa.us> Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="=-tyawsElsu3INch108LXq" X-Mailer: Ximian Evolution 1.0.8 Date: 26 Sep 2002 12:36:57 -0500 Message-Id: <1033061818.23390.12.camel@mouse.copelandconsulting.net> Mime-Version: 1.0 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1208 X-Sequence-Number: 30789 --=-tyawsElsu3INch108LXq Content-Type: text/plain Content-Transfer-Encoding: quoted-printable On Thu, 2002-09-26 at 11:41, Bruce Momjian wrote: > Shridhar Daithankar wrote: > > I might have found the bottleneck, although by accident. Mysql was runn= ing out=20 > > of space while creating index. So my friend shut down mysql and tried t= o move=20 > > things by hand to create links. He noticed that even things like cp wer= e=20 > > terribly slow and it hit us.. May be the culprit is the file system. Ex= t3 in=20 > > this case.=20 >=20 > I just added a file system and multi-cpu section to my performance > tuning paper: >=20 > http://www.ca.postgresql.org/docs/momjian/hw_performance/ >=20 > The paper does recommend ext3, but the differences between file systems > are very small. If you are seeing 'cp' as slow, I wonder if it may be > something more general, like poorly tuned hardware or something. You can > use 'dd' to throw some data around the file system and see if that is > showing slowness; compare those numbers to another machine that has > different hardware/OS. That's a good point. Also, if you're using IDE, you do need to verify that you're using DMA and proper PIO mode if at possible. Also, big performance improvements can be seen by making sure your IDE bus speed has been properly configured. The drivetweak-gtk and hdparm utilities can make huge difference in performance. Just be sure you know what the heck your doing when you mess with those. Greg --=-tyawsElsu3INch108LXq Content-Type: application/pgp-signature; name=signature.asc Content-Description: This is a digitally signed message part -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.7 (GNU/Linux) iD8DBQA9k0W54lr1bpbcL6kRAsTSAJ410S530QfaeTjTxEaICSnzhUbNOwCeNGvr tTGuXDVz190FH55un7vEjrc= =HFjK -----END PGP SIGNATURE----- --=-tyawsElsu3INch108LXq-- From pgsql-general-owner@postgresql.org Thu Sep 26 14:13:31 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 27240476DA8; Thu, 26 Sep 2002 13:45:12 -0400 (EDT) Received: from CopelandConsulting.Net (dsl-24293-ld.customer.centurytel.net [209.142.135.135]) by postgresql.org (Postfix) with ESMTP id D8262476D8F; Thu, 26 Sep 2002 13:45:10 -0400 (EDT) Received: from mouse.copelandconsulting.net (mouse.copelandconsulting.net [192.168.1.2]) by CopelandConsulting.Net (8.10.1/8.10.1) with ESMTP id g8QHiGu14092; Thu, 26 Sep 2002 12:44:16 -0500 (CDT) X-Trade-Id: To: Bruce Momjian Cc: shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List In-Reply-To: <200209261641.g8QGfYc04099@candle.pha.pa.us> References: <200209261641.g8QGfYc04099@candle.pha.pa.us> Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="=-6mav3WK9RCaVOqe8tRq/" X-Mailer: Ximian Evolution 1.0.8 Date: 26 Sep 2002 12:44:22 -0500 Message-Id: <1033062262.23475.16.camel@mouse.copelandconsulting.net> Mime-Version: 1.0 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1207 X-Sequence-Number: 30788 --=-6mav3WK9RCaVOqe8tRq/ Content-Type: text/plain Content-Transfer-Encoding: quoted-printable On Thu, 2002-09-26 at 11:41, Bruce Momjian wrote: > Shridhar Daithankar wrote: > > I might have found the bottleneck, although by accident. Mysql was runn= ing out=20 > > of space while creating index. So my friend shut down mysql and tried t= o move=20 > > things by hand to create links. He noticed that even things like cp wer= e=20 > > terribly slow and it hit us.. May be the culprit is the file system. Ex= t3 in=20 > > this case.=20 >=20 > I just added a file system and multi-cpu section to my performance > tuning paper: >=20 > http://www.ca.postgresql.org/docs/momjian/hw_performance/ >=20 > The paper does recommend ext3, but the differences between file systems > are very small. If you are seeing 'cp' as slow, I wonder if it may be > something more general, like poorly tuned hardware or something. You can > use 'dd' to throw some data around the file system and see if that is > showing slowness; compare those numbers to another machine that has > different hardware/OS. >=20 > Also, though ext3 is slower, turning fsync off should make ext3 function > similar to ext2. That would be an interesting test if you suspect ext3. I'm curious as to why you recommended ext3 versus some other (JFS, XFS). Do you have tests which validate that recommendation or was it a simple matter of getting the warm fuzzies from familiarity? Greg --=-6mav3WK9RCaVOqe8tRq/ Content-Type: application/pgp-signature; name=signature.asc Content-Description: This is a digitally signed message part -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.7 (GNU/Linux) iD8DBQA9k0d24lr1bpbcL6kRAkGfAJ0en60jxkx1LsCX8HIzsjHgA8MnKQCffW/S m+nGg6nihDZ/JABT4dNcuGo= =Itl1 -----END PGP SIGNATURE----- --=-6mav3WK9RCaVOqe8tRq/-- From pgsql-general-owner@postgresql.org Thu Sep 26 14:46:12 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 583924762C6; Thu, 26 Sep 2002 14:46:11 -0400 (EDT) Received: from mail1.ihs.com (mail1.ihs.com [170.207.70.222]) by postgresql.org (Postfix) with ESMTP id 9179947606A; Thu, 26 Sep 2002 14:46:09 -0400 (EDT) Received: from css120.ihs.com (css120.ihs.com [170.207.105.120]) by mail1.ihs.com (8.12.6/8.12.6) with ESMTP id g8QIjU0P015223; Thu, 26 Sep 2002 12:45:30 -0600 (MDT) Date: Thu, 26 Sep 2002 12:41:55 -0600 (MDT) From: "scott.marlowe" To: Shridhar Daithankar Cc: , Subject: Re: [HACKERS] Performance while loading data and indexing In-Reply-To: <3D931882.31859.134B9E4C@localhost> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-MailBodyFilter: Message body has not been filtered X-MailScanner: Found to be clean X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1209 X-Sequence-Number: 30790 If you are seeing very slow performance on a drive set, check dmesg to see if you're getting SCSI bus errors or something similar. If your drives aren't properly terminated then the performance will suffer a great deal. From pgsql-general-owner@postgresql.org Thu Sep 26 16:01:20 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 09482476052; Thu, 26 Sep 2002 16:01:18 -0400 (EDT) Received: from candle.pha.pa.us (momjian.navpoint.com [207.106.42.251]) by postgresql.org (Postfix) with ESMTP id A0F8D474E5C; Thu, 26 Sep 2002 16:01:14 -0400 (EDT) Received: (from pgman@localhost) by candle.pha.pa.us (8.11.6/8.10.1) id g8QK0mG10553; Thu, 26 Sep 2002 16:00:48 -0400 (EDT) From: Bruce Momjian Message-Id: <200209262000.g8QK0mG10553@candle.pha.pa.us> Subject: Re: [HACKERS] Performance while loading data and indexing In-Reply-To: <1033062262.23475.16.camel@mouse.copelandconsulting.net> To: Greg Copeland Date: Thu, 26 Sep 2002 16:00:48 -0400 (EDT) Cc: shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List X-Mailer: ELM [version 2.4ME+ PL99 (25)] MIME-Version: 1.0 Content-Type: multipart/mixed; boundary=ELM1033070448-26881-0_ Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1211 X-Sequence-Number: 30792 --ELM1033070448-26881-0_ Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII Greg Copeland wrote: > > The paper does recommend ext3, but the differences between file systems > > are very small. If you are seeing 'cp' as slow, I wonder if it may be > > something more general, like poorly tuned hardware or something. You can > > use 'dd' to throw some data around the file system and see if that is > > showing slowness; compare those numbers to another machine that has > > different hardware/OS. > > > > Also, though ext3 is slower, turning fsync off should make ext3 function > > similar to ext2. That would be an interesting test if you suspect ext3. > > I'm curious as to why you recommended ext3 versus some other (JFS, > XFS). Do you have tests which validate that recommendation or was it a > simple matter of getting the warm fuzzies from familiarity? I used the attached email as a reference. I just changed the wording to be: File system choice is particularly difficult on Linux because there are so many file system choices, and none of them are optimal: ext2 is not entirely crash-safe, ext3 and xfs are journal-based, and Reiser is optimized for small files. Fortunately, the journaling file systems aren't significantly slower than ext2 so they are probably the best choice. so I don't specifically recommend ext3 anymore. As I remember, ext3 is good only in that it can read ext2 file systems. I think XFS may be the best bet. Can anyone clarify if "data=writeback" is safe for PostgreSQL. Specifically, are the data files recovered properly or is this option only for a filesystem containing WAL? -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073 --ELM1033070448-26881-0_ Content-Transfer-Encoding: 7bit Content-Type: text/plain Content-Disposition: inline; filename="/bjm/perf" --ELM1033070448-26881-0_-- From pgsql-general-owner@postgresql.org Thu Sep 26 16:42:13 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 5B9B147676D; Thu, 26 Sep 2002 16:42:12 -0400 (EDT) Received: from bob.samurai.com (bob.samurai.com [205.207.28.75]) by postgresql.org (Postfix) with ESMTP id E1D784762B7; Thu, 26 Sep 2002 16:42:11 -0400 (EDT) Received: from boston.samurai.com (DU179.N224.ResNet.QueensU.CA [130.15.224.179]) by bob.samurai.com (Postfix) with ESMTP id E9B1A1EAC; Thu, 26 Sep 2002 16:42:10 -0400 (EDT) To: Bruce Momjian Cc: shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: [HACKERS] Performance while loading data and indexing References: <200209261641.g8QGfYc04099@candle.pha.pa.us> From: Neil Conway In-Reply-To: <200209261641.g8QGfYc04099@candle.pha.pa.us> Date: 26 Sep 2002 16:41:49 -0400 Message-ID: <871y7g1o1e.fsf@mailbox.samurai.com> Lines: 20 User-Agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.2 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1212 X-Sequence-Number: 30793 Bruce Momjian writes: > The paper does recommend ext3, but the differences between file systems > are very small. Well, I only did a very rough benchmark (a few runs of pgbench), but the results I found were drastically different: ext2 was significantly faster (~50%) than ext3-writeback, which was in turn significantly faster (~25%) than ext3-ordered. > Also, though ext3 is slower, turning fsync off should make ext3 function > similar to ext2. Why would that be? Cheers, Neil -- Neil Conway || PGP Key ID: DB3C29FC From pgsql-general-owner@postgresql.org Thu Sep 26 16:46:10 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id AD961476147; Thu, 26 Sep 2002 16:46:07 -0400 (EDT) Received: from candle.pha.pa.us (momjian.navpoint.com [207.106.42.251]) by postgresql.org (Postfix) with ESMTP id C734B476052; Thu, 26 Sep 2002 16:46:05 -0400 (EDT) Received: (from pgman@localhost) by candle.pha.pa.us (8.11.6/8.10.1) id g8QKjtv21744; Thu, 26 Sep 2002 16:45:55 -0400 (EDT) From: Bruce Momjian Message-Id: <200209262045.g8QKjtv21744@candle.pha.pa.us> Subject: Re: [HACKERS] Performance while loading data and indexing In-Reply-To: <871y7g1o1e.fsf@mailbox.samurai.com> To: Neil Conway Date: Thu, 26 Sep 2002 16:45:54 -0400 (EDT) Cc: shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org X-Mailer: ELM [version 2.4ME+ PL99 (25)] MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1213 X-Sequence-Number: 30794 Neil Conway wrote: > Bruce Momjian writes: > > The paper does recommend ext3, but the differences between file systems > > are very small. > > Well, I only did a very rough benchmark (a few runs of pgbench), but > the results I found were drastically different: ext2 was significantly > faster (~50%) than ext3-writeback, which was in turn significantly > faster (~25%) than ext3-ordered. Wow. That leaves no good Linux file system alternatives. PostgreSQL just wants an ordinary file system that has reliable recovery from a crash. > > Also, though ext3 is slower, turning fsync off should make ext3 function > > similar to ext2. > > Why would that be? I assumed it was the double fsync for the normal and journal that made the journalling file systems slog. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073 From pgsql-hackers-owner@postgresql.org Thu Sep 26 16:50:43 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 59D4D476083 for ; Thu, 26 Sep 2002 16:50:42 -0400 (EDT) Received: from bachata.cybertec.at (unknown [62.116.21.146]) by postgresql.org (Postfix) with SMTP id 1856E474E5C for ; Thu, 26 Sep 2002 16:50:41 -0400 (EDT) Received: (qmail 26347 invoked from network); 26 Sep 2002 20:50:47 -0000 Received: from unknown (HELO cybertec.at) (62.116.21.147) by 62.116.21.146 with SMTP; 26 Sep 2002 20:50:47 -0000 Message-ID: <3D937442.40902@cybertec.at> Date: Thu, 26 Sep 2002 22:55:30 +0200 From: =?ISO-8859-1?Q?Hans-J=FCrgen_Sch=F6nig?= Reply-To: hs@cybertec.at User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.1) Gecko/20020827 X-Accept-Language: en-us, en MIME-Version: 1.0 To: Neil Conway , pgsql-hackers@postgresql.org Subject: Re: [GENERAL] Performance while loading data and indexing References: <200209261641.g8QGfYc04099@candle.pha.pa.us> <871y7g1o1e.fsf@mailbox.samurai.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1576 X-Sequence-Number: 29533 I have seen various benchmarks where XFS seems to perform best when it comes to huge amounts of data and many files (due to balanced internal b+ trees). also, XFS seems to be VERY mature and very stable. ext2/3 don't seem to be that fast in most of the benchmarks. i did some testing with reiser some time ago. the problem is that it seems to restore a very historic consistent snapshot of the data. XFS seems to be much better in this respect. i have not tested JFS yet (but on this damn AIX beside me) from my point of view i strongly recommend XFS (maybe somebody from RedHat should think about it). Hans Neil Conway wrote: >Bruce Momjian writes: > > >>The paper does recommend ext3, but the differences between file systems >>are very small. >> >> > >Well, I only did a very rough benchmark (a few runs of pgbench), but >the results I found were drastically different: ext2 was significantly >faster (~50%) than ext3-writeback, which was in turn significantly >faster (~25%) than ext3-ordered. > > > >>Also, though ext3 is slower, turning fsync off should make ext3 function >>similar to ext2. >> >> > >Why would that be? > >Cheers, > >Neil > > > -- *Cybertec Geschwinde u Schoenig* Ludo-Hartmannplatz 1/14, A-1160 Vienna, Austria Tel: +43/1/913 68 09; +43/664/233 90 75 www.postgresql.at , cluster.postgresql.at , www.cybertec.at , kernel.cybertec.at From pgsql-general-owner@postgresql.org Thu Sep 26 16:57:22 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id D2740476F59; Thu, 26 Sep 2002 16:57:21 -0400 (EDT) Received: from candle.pha.pa.us (momjian.navpoint.com [207.106.42.251]) by postgresql.org (Postfix) with ESMTP id AF1B947702E; Thu, 26 Sep 2002 16:57:14 -0400 (EDT) Received: (from pgman@localhost) by candle.pha.pa.us (8.11.6/8.10.1) id g8QKv3Z22867; Thu, 26 Sep 2002 16:57:03 -0400 (EDT) From: Bruce Momjian Message-Id: <200209262057.g8QKv3Z22867@candle.pha.pa.us> Subject: Re: [HACKERS] Performance while loading data and indexing In-Reply-To: <871y7g1o1e.fsf@mailbox.samurai.com> To: Neil Conway Date: Thu, 26 Sep 2002 16:57:03 -0400 (EDT) Cc: shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org X-Mailer: ELM [version 2.4ME+ PL99 (25)] MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1215 X-Sequence-Number: 30796 Neil Conway wrote: > Bruce Momjian writes: > > The paper does recommend ext3, but the differences between file systems > > are very small. > > Well, I only did a very rough benchmark (a few runs of pgbench), but > the results I found were drastically different: ext2 was significantly > faster (~50%) than ext3-writeback, which was in turn significantly > faster (~25%) than ext3-ordered. > > > Also, though ext3 is slower, turning fsync off should make ext3 function > > similar to ext2. > > Why would that be? OK, I changed the text to: File system choice is particularly difficult on Linux because there are so many file system choices, and none of them are optimal: ext2 is not entirely crash-safe, ext3, xfs, and jfs are journal-based, and Reiser is optimized for small files and does journalling. The journalling file systems can be significantly slower than ext2 but when crash recovery is required, ext2 isn't an option. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073 From pgsql-general-owner@postgresql.org Thu Sep 26 17:03:41 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 8D545476FB9; Thu, 26 Sep 2002 17:03:40 -0400 (EDT) Received: from bob.samurai.com (bob.samurai.com [205.207.28.75]) by postgresql.org (Postfix) with ESMTP id 1A27A476F86; Thu, 26 Sep 2002 17:03:40 -0400 (EDT) Received: from boston.samurai.com (DU179.N224.ResNet.QueensU.CA [130.15.224.179]) by bob.samurai.com (Postfix) with ESMTP id 0015F1EAB; Thu, 26 Sep 2002 17:03:39 -0400 (EDT) To: Bruce Momjian Cc: Neil Conway , shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: [HACKERS] Performance while loading data and indexing References: <200209262045.g8QKjtv21744@candle.pha.pa.us> From: Neil Conway In-Reply-To: <200209262045.g8QKjtv21744@candle.pha.pa.us> Date: 26 Sep 2002 17:03:26 -0400 Message-ID: <87vg4szco1.fsf@mailbox.samurai.com> Lines: 29 User-Agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.2 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1217 X-Sequence-Number: 30798 Bruce Momjian writes: > Wow. That leaves no good Linux file system alternatives. > PostgreSQL just wants an ordinary file system that has reliable > recovery from a crash. I'm not really familiar with the reasoning behind ext2's reputation as recovering poorly from crashes; if we fsync a WAL record to disk before we lose power, can't we recover reliably, even with ext2? > > > Also, though ext3 is slower, turning fsync off should make ext3 > > > function similar to ext2. > > > > Why would that be? > > I assumed it was the double fsync for the normal and journal that > made the journalling file systems slog. Well, a journalling file system would need to write a journal entry and flush that to disk, even if fsync is disabled -- whereas without fsync enabled, ext2 doesn't have to flush anything to disk. ISTM that the performance advantage of ext2 over ext3 is should be even larger when fsync is not enabled. Cheers, Neil -- Neil Conway || PGP Key ID: DB3C29FC From pgsql-hackers-owner@postgresql.org Thu Sep 26 17:04:07 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id CF22C477069 for ; Thu, 26 Sep 2002 17:04:06 -0400 (EDT) Received: from CopelandConsulting.Net (dsl-24293-ld.customer.centurytel.net [209.142.135.135]) by postgresql.org (Postfix) with ESMTP id 9BEAA476C0B for ; Thu, 26 Sep 2002 17:03:54 -0400 (EDT) Received: from mouse.copelandconsulting.net (mouse.copelandconsulting.net [192.168.1.2]) by CopelandConsulting.Net (8.10.1/8.10.1) with ESMTP id g8QL3hu15120; Thu, 26 Sep 2002 16:03:44 -0500 (CDT) X-Trade-Id: To: hs@cybertec.at Cc: PostgresSQL Hackers Mailing List In-Reply-To: <3D937442.40902@cybertec.at> References: <200209261641.g8QGfYc04099@candle.pha.pa.us> <871y7g1o1e.fsf@mailbox.samurai.com> <3D937442.40902@cybertec.at> Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="=-ZqXCINKzOxv8OZvLp8ID" X-Mailer: Ximian Evolution 1.0.8 Date: 26 Sep 2002 16:03:51 -0500 Message-Id: <1033074232.23474.42.camel@mouse.copelandconsulting.net> Mime-Version: 1.0 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1582 X-Sequence-Number: 29539 --=-ZqXCINKzOxv8OZvLp8ID Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable I tend to agree with this though I have nothing to back up it with. My impression is that XFS does very well for large files. Accepting that as fact?, my impression is that XFS historically does well for database's. Again, I have nothing to back that up other than hear-say and conjecture. Greg On Thu, 2002-09-26 at 15:55, Hans-J=FCrgen Sch=F6nig wrote: > I have seen various benchmarks where XFS seems to perform best when it=20 > comes to huge amounts of data and many files (due to balanced internal=20 > b+ trees). > also, XFS seems to be VERY mature and very stable. > ext2/3 don't seem to be that fast in most of the benchmarks. >=20 > i did some testing with reiser some time ago. the problem is that it=20 > seems to restore a very historic consistent snapshot of the data. XFS=20 > seems to be much better in this respect. >=20 > i have not tested JFS yet (but on this damn AIX beside me) > from my point of view i strongly recommend XFS (maybe somebody from=20 > RedHat should think about it). >=20 > Hans >=20 >=20 > Neil Conway wrote: >=20 > >Bruce Momjian writes: > >=20=20 > > > >>The paper does recommend ext3, but the differences between file systems > >>are very small. > >>=20=20=20=20 > >> > > > >Well, I only did a very rough benchmark (a few runs of pgbench), but > >the results I found were drastically different: ext2 was significantly > >faster (~50%) than ext3-writeback, which was in turn significantly > >faster (~25%) than ext3-ordered. > > > >=20=20 > > > >>Also, though ext3 is slower, turning fsync off should make ext3 function > >>similar to ext2. > >>=20=20=20=20 > >> > > > >Why would that be? > > > >Cheers, > > > >Neil > > > >=20=20 > > >=20 >=20 > --=20 > *Cybertec Geschwinde u Schoenig* > Ludo-Hartmannplatz 1/14, A-1160 Vienna, Austria > Tel: +43/1/913 68 09; +43/664/233 90 75 > www.postgresql.at , cluster.postgresql.at=20 > , www.cybertec.at=20 > , kernel.cybertec.at >=20 >=20 > ---------------------------(end of broadcast)--------------------------- > TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org --=-ZqXCINKzOxv8OZvLp8ID Content-Type: application/pgp-signature; name=signature.asc Content-Description: This is a digitally signed message part -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.7 (GNU/Linux) iD8DBQA9k3Y34lr1bpbcL6kRAlvSAJ9DicilkkEypigomt/wfiO5nHyqqQCeOTHL krIfkgUxrfgr50wh8oI93Lc= =pUiE -----END PGP SIGNATURE----- --=-ZqXCINKzOxv8OZvLp8ID-- From pgsql-hackers-owner@postgresql.org Sat Sep 28 13:29:40 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id A7844476EEB; Thu, 26 Sep 2002 17:01:28 -0400 (EDT) Received: from gatekeeper.d2000.com (gatekeeper.d2000.com [208.32.117.78]) by postgresql.org (Postfix) with ESMTP id AE297476E66; Thu, 26 Sep 2002 17:01:27 -0400 (EDT) Received: (from bin@localhost) by gatekeeper.d2000.com (8.9.3/8.7.3) id QAA41714; Thu, 26 Sep 2002 16:01:19 -0500 (CDT) X-Authentication-Warning: gatekeeper.d2000.com: bin set sender to using -f Received: from (nt.d2000.com [205.164.66.20]) by gatekeeper.d2000.com via smap (V2.1) id xma041690; Thu, 26 Sep 02 16:00:47 -0500 Received: from xl.d2000.com (unverified) by nt.d2000.com (Content Technologies SMTPRS 4.1.5) with ESMTP id ; Thu, 26 Sep 2002 16:00:45 -0500 Received: from materialpcow5r (backup.d2000.com [205.164.66.13]) by xl.d2000.com (8.8.8/8.7.3) with SMTP id QAA05701; Thu, 26 Sep 2002 16:00:33 -0500 (CDT) From: "James Maes" To: "Bruce Momjian" , "Neil Conway" Cc: , , Subject: Re: [GENERAL] Performance while loading data and indexing Date: Thu, 26 Sep 2002 16:06:07 -0500 Message-ID: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit X-Priority: 3 (Normal) X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook IMO, Build 9.0.2416 (9.0.2910.0) X-MimeOLE: Produced By Microsoft MimeOLE V5.50.4807.1700 Importance: Normal In-Reply-To: <200209262057.g8QKv3Z22867@candle.pha.pa.us> X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1669 X-Sequence-Number: 29626 Has there been any thought of providing RAW disk support to bypass the fs? -----Original Message----- From: pgsql-hackers-owner@postgresql.org [mailto:pgsql-hackers-owner@postgresql.org]On Behalf Of Bruce Momjian Sent: Thursday, September 26, 2002 3:57 PM To: Neil Conway Cc: shridhar_daithankar@persistent.co.in; pgsql-hackers@postgresql.org; pgsql-general@postgresql.org Subject: Re: [HACKERS] [GENERAL] Performance while loading data and indexing Neil Conway wrote: > Bruce Momjian writes: > > The paper does recommend ext3, but the differences between file systems > > are very small. > > Well, I only did a very rough benchmark (a few runs of pgbench), but > the results I found were drastically different: ext2 was significantly > faster (~50%) than ext3-writeback, which was in turn significantly > faster (~25%) than ext3-ordered. > > > Also, though ext3 is slower, turning fsync off should make ext3 function > > similar to ext2. > > Why would that be? OK, I changed the text to: File system choice is particularly difficult on Linux because there are so many file system choices, and none of them are optimal: ext2 is not entirely crash-safe, ext3, xfs, and jfs are journal-based, and Reiser is optimized for small files and does journalling. The journalling file systems can be significantly slower than ext2 but when crash recovery is required, ext2 isn't an option. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073 ---------------------------(end of broadcast)--------------------------- TIP 5: Have you checked our extensive FAQ? http://www.postgresql.org/users-lounge/docs/faq.html From pgsql-general-owner@postgresql.org Thu Sep 26 17:08:13 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 3B9ED4769E6; Thu, 26 Sep 2002 17:08:13 -0400 (EDT) Received: from candle.pha.pa.us (momjian.navpoint.com [207.106.42.251]) by postgresql.org (Postfix) with ESMTP id 2AD5C476241; Thu, 26 Sep 2002 17:08:11 -0400 (EDT) Received: (from pgman@localhost) by candle.pha.pa.us (8.11.6/8.10.1) id g8QL7vN25965; Thu, 26 Sep 2002 17:07:57 -0400 (EDT) From: Bruce Momjian Message-Id: <200209262107.g8QL7vN25965@candle.pha.pa.us> Subject: Re: [HACKERS] Performance while loading data and indexing In-Reply-To: <87vg4szco1.fsf@mailbox.samurai.com> To: Neil Conway Date: Thu, 26 Sep 2002 17:07:57 -0400 (EDT) Cc: shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org X-Mailer: ELM [version 2.4ME+ PL99 (25)] MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1218 X-Sequence-Number: 30799 Neil Conway wrote: > Bruce Momjian writes: > > Wow. That leaves no good Linux file system alternatives. > > PostgreSQL just wants an ordinary file system that has reliable > > recovery from a crash. > > I'm not really familiar with the reasoning behind ext2's reputation as > recovering poorly from crashes; if we fsync a WAL record to disk > before we lose power, can't we recover reliably, even with ext2? > > > > > Also, though ext3 is slower, turning fsync off should make ext3 > > > > function similar to ext2. > > > > > > Why would that be? > > > > I assumed it was the double fsync for the normal and journal that > > made the journalling file systems slog. > > Well, a journalling file system would need to write a journal entry > and flush that to disk, even if fsync is disabled -- whereas without > fsync enabled, ext2 doesn't have to flush anything to disk. ISTM that > the performance advantage of ext2 over ext3 is should be even larger > when fsync is not enabled. Yes, it is still double-writing. I just thought that if that wasn't happening while the db was waiting for a commit that it wouldn't be too bad. Is it just me or do all the Linux file systems seem like they are lacking something when PostgreSQL is concerned? We just want a UFS-like file system on Linux and no one has it. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073 From pgsql-general-owner@postgresql.org Thu Sep 26 17:09:32 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 4AD7A476FCC; Thu, 26 Sep 2002 17:09:31 -0400 (EDT) Received: from CopelandConsulting.Net (dsl-24293-ld.customer.centurytel.net [209.142.135.135]) by postgresql.org (Postfix) with ESMTP id 22CE6476FC8; Thu, 26 Sep 2002 17:09:30 -0400 (EDT) Received: from mouse.copelandconsulting.net (mouse.copelandconsulting.net [192.168.1.2]) by CopelandConsulting.Net (8.10.1/8.10.1) with ESMTP id g8QL97u03616; Thu, 26 Sep 2002 16:09:07 -0500 (CDT) X-Trade-Id: To: Neil Conway Cc: Bruce Momjian , shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List In-Reply-To: <87vg4szco1.fsf@mailbox.samurai.com> References: <200209262045.g8QKjtv21744@candle.pha.pa.us> <87vg4szco1.fsf@mailbox.samurai.com> Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="=-ukBN/VLqUCvLMvEd/iTv" X-Mailer: Ximian Evolution 1.0.8 Date: 26 Sep 2002 16:09:15 -0500 Message-Id: <1033074555.23344.48.camel@mouse.copelandconsulting.net> Mime-Version: 1.0 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1219 X-Sequence-Number: 30800 --=-ukBN/VLqUCvLMvEd/iTv Content-Type: text/plain Content-Transfer-Encoding: quoted-printable On Thu, 2002-09-26 at 16:03, Neil Conway wrote: > Bruce Momjian writes: > > Wow. That leaves no good Linux file system alternatives. > > PostgreSQL just wants an ordinary file system that has reliable > > recovery from a crash. >=20 > I'm not really familiar with the reasoning behind ext2's reputation as > recovering poorly from crashes; if we fsync a WAL record to disk > before we lose power, can't we recover reliably, even with ext2? Well, I have experienced data loss from ext2 before. Also, recovery from crashes on large file systems take a very, very long time. I can't imagine anyone running a production database on an ext2 file system having 10's or even 100's of GB. Ouch. Recovery would take forever!=20 Even recovery on small file systems (2-8G) can take extended periods of time. Especially so on IDE systems. Even then manual intervention is not uncommon. While I can't say that x, y or z is the best FS to use on Linux, I can say that ext2 is probably an exceptionally poor choice from a reliability and/or uptime perspective. Greg --=-ukBN/VLqUCvLMvEd/iTv Content-Type: application/pgp-signature; name=signature.asc Content-Description: This is a digitally signed message part -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.7 (GNU/Linux) iD8DBQA9k3d74lr1bpbcL6kRAttbAJ44dhAlrsYjtTfGuieCrbJBqLV7PwCfWElV QDa/ABmzxCPU/REOuseR7bo= =Sg1u -----END PGP SIGNATURE----- --=-ukBN/VLqUCvLMvEd/iTv-- From pgsql-general-owner@postgresql.org Thu Sep 26 17:17:43 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 4E5B1474E5C; Thu, 26 Sep 2002 17:17:42 -0400 (EDT) Received: from bob.samurai.com (bob.samurai.com [205.207.28.75]) by postgresql.org (Postfix) with ESMTP id 3C927476F34; Thu, 26 Sep 2002 17:17:41 -0400 (EDT) Received: from boston.samurai.com (DU179.N224.ResNet.QueensU.CA [130.15.224.179]) by bob.samurai.com (Postfix) with ESMTP id 066721ECE; Thu, 26 Sep 2002 17:17:42 -0400 (EDT) To: Greg Copeland Cc: Neil Conway , Bruce Momjian , shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List Subject: Re: [HACKERS] Performance while loading data and indexing References: <200209262045.g8QKjtv21744@candle.pha.pa.us> <87vg4szco1.fsf@mailbox.samurai.com> <1033074555.23344.48.camel@mouse.copelandconsulting.net> From: Neil Conway In-Reply-To: <1033074555.23344.48.camel@mouse.copelandconsulting.net> Date: 26 Sep 2002 17:17:30 -0400 Message-ID: <87n0q4zc0l.fsf@mailbox.samurai.com> Lines: 24 User-Agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.2 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1220 X-Sequence-Number: 30801 Greg Copeland writes: > On Thu, 2002-09-26 at 16:03, Neil Conway wrote: > > I'm not really familiar with the reasoning behind ext2's > > reputation as recovering poorly from crashes; if we fsync a WAL > > record to disk before we lose power, can't we recover reliably, > > even with ext2? > > Well, I have experienced data loss from ext2 before. Also, recovery > from crashes on large file systems take a very, very long time. Yes, but wouldn't you face exactly the same issues if you ran a UFS-like filesystem in asynchronous mode? Albeit it's not the default, but performance in synchronous mode is usually pretty poor. The fact that ext2 defaults to asynchronous mode and UFS (at least on the BSDs) defaults to synchronous mode seems like a total non-issue to me. Is there any more to the alleged difference in reliability? Cheers, Neil -- Neil Conway || PGP Key ID: DB3C29FC From pgsql-general-owner@postgresql.org Thu Sep 26 17:32:17 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 938DE4769BA; Thu, 26 Sep 2002 17:32:15 -0400 (EDT) Received: from varsoon.wireboard.com (www.wireboard.com [216.151.155.101]) by postgresql.org (Postfix) with ESMTP id 0DC964769AE; Thu, 26 Sep 2002 17:32:15 -0400 (EDT) Received: from doug by varsoon.wireboard.com with local (Exim 3.35 #1) id 17ugEl-0006nF-00; Thu, 26 Sep 2002 17:31:55 -0400 To: Bruce Momjian From: Doug McNaught Cc: Greg Copeland , shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List Subject: Re: [HACKERS] Performance while loading data and indexing References: <200209262000.g8QK0mG10553@candle.pha.pa.us> Date: 26 Sep 2002 17:31:55 -0400 In-Reply-To: Bruce Momjian's message of "Thu, 26 Sep 2002 16:00:48 -0400 (EDT)" Message-ID: Lines: 24 User-Agent: Gnus/5.0806 (Gnus v5.8.6) Emacs/20.7 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1221 X-Sequence-Number: 30802 Bruce Momjian writes: > Can anyone clarify if "data=writeback" is safe for PostgreSQL. > Specifically, are the data files recovered properly or is this option > only for a filesystem containing WAL? "data=writeback" means that no data is journaled, just metadata (which is like XFS or Reiser). An fsync() call should still do what it normally does, commit the writes to disk before returning. "data=journal" journals all data and is the slowest and safest. "data=ordered" writes out data blocks before committing a journal transaction, which is faster than full data journaling (since data doesn't get written twice) and almost as safe. "data=writeback" is noted to keep obsolete data in the case of some crashes (since the data may not have been written yet) but a completed fsync() should ensure that the data is valid. So I guess I'd probably use data=ordered for an all-on-one-fs installation, and data=writeback for a WAL-only drive. Hope this helps... -Doug From pgsql-general-owner@postgresql.org Thu Sep 26 17:32:39 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 9C0C4477044 for ; Thu, 26 Sep 2002 17:32:38 -0400 (EDT) Received: from sss.pgh.pa.us (unknown [192.204.191.242]) by postgresql.org (Postfix) with ESMTP id 591CA476F89 for ; Thu, 26 Sep 2002 17:32:26 -0400 (EDT) Received: from sss2.sss.pgh.pa.us (tgl@localhost [127.0.0.1]) by sss.pgh.pa.us (8.12.5/8.12.5) with ESMTP id g8QLW1hR012931; Thu, 26 Sep 2002 17:32:01 -0400 (EDT) To: Neil Conway Cc: Bruce Momjian , shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: [HACKERS] Performance while loading data and indexing In-reply-to: <87vg4szco1.fsf@mailbox.samurai.com> References: <200209262045.g8QKjtv21744@candle.pha.pa.us> <87vg4szco1.fsf@mailbox.samurai.com> Comments: In-reply-to Neil Conway message dated "26 Sep 2002 17:03:26 -0400" Date: Thu, 26 Sep 2002 17:32:01 -0400 Message-ID: <12930.1033075921@sss.pgh.pa.us> From: Tom Lane X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1222 X-Sequence-Number: 30803 Neil Conway writes: > I'm not really familiar with the reasoning behind ext2's reputation as > recovering poorly from crashes; if we fsync a WAL record to disk > before we lose power, can't we recover reliably, even with ext2? Up to a point. We do assume that the filesystem won't lose checkpointed (sync'd) writes to data files. To the extent that the filesystem is vulnerable to corruption of its own metadata for a file (indirect blocks or whatever ext2 uses), that's not a completely safe assumption. We'd be happiest with a filesystem that journals its own metadata and not the user data in the file(s). I dunno if there are any. Hmm, maybe this is why Oracle likes doing their own filesystem on a raw device... regards, tom lane From pgsql-general-owner@postgresql.org Thu Sep 26 17:37:39 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id F267E476FCC; Thu, 26 Sep 2002 17:37:38 -0400 (EDT) Received: from varsoon.wireboard.com (www.wireboard.com [216.151.155.101]) by postgresql.org (Postfix) with ESMTP id 1DF13476FC3; Thu, 26 Sep 2002 17:37:38 -0400 (EDT) Received: from doug by varsoon.wireboard.com with local (Exim 3.35 #1) id 17ugJq-0006nf-00; Thu, 26 Sep 2002 17:37:10 -0400 To: Tom Lane From: Doug McNaught Cc: Neil Conway , Bruce Momjian , shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: [HACKERS] Performance while loading data and indexing References: <200209262045.g8QKjtv21744@candle.pha.pa.us> <87vg4szco1.fsf@mailbox.samurai.com> <12930.1033075921@sss.pgh.pa.us> Date: 26 Sep 2002 17:37:10 -0400 In-Reply-To: Tom Lane's message of "Thu, 26 Sep 2002 17:32:01 -0400" Message-ID: Lines: 8 User-Agent: Gnus/5.0806 (Gnus v5.8.6) Emacs/20.7 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1223 X-Sequence-Number: 30804 Tom Lane writes: > We'd be happiest with a filesystem that journals its own metadata and > not the user data in the file(s). I dunno if there are any. ext3 with data=writeback? (See my previous message to Bruce). -Doug From pgsql-general-owner@postgresql.org Thu Sep 26 17:39:35 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 0C3B54760BD; Thu, 26 Sep 2002 17:39:35 -0400 (EDT) Received: from candle.pha.pa.us (momjian.navpoint.com [207.106.42.251]) by postgresql.org (Postfix) with ESMTP id 7E8AC4770C8; Thu, 26 Sep 2002 17:39:31 -0400 (EDT) Received: (from pgman@localhost) by candle.pha.pa.us (8.11.6/8.10.1) id g8QLdEE08861; Thu, 26 Sep 2002 17:39:14 -0400 (EDT) From: Bruce Momjian Message-Id: <200209262139.g8QLdEE08861@candle.pha.pa.us> Subject: Re: [HACKERS] Performance while loading data and indexing In-Reply-To: <87n0q4zc0l.fsf@mailbox.samurai.com> To: Neil Conway Date: Thu, 26 Sep 2002 17:39:14 -0400 (EDT) Cc: Greg Copeland , shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List X-Mailer: ELM [version 2.4ME+ PL99 (25)] MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1224 X-Sequence-Number: 30805 Neil Conway wrote: > Greg Copeland writes: > > On Thu, 2002-09-26 at 16:03, Neil Conway wrote: > > > I'm not really familiar with the reasoning behind ext2's > > > reputation as recovering poorly from crashes; if we fsync a WAL > > > record to disk before we lose power, can't we recover reliably, > > > even with ext2? > > > > Well, I have experienced data loss from ext2 before. Also, recovery > > from crashes on large file systems take a very, very long time. > > Yes, but wouldn't you face exactly the same issues if you ran a > UFS-like filesystem in asynchronous mode? Albeit it's not the default, > but performance in synchronous mode is usually pretty poor. Yes, before UFS had soft updates, the synchronous nature of UFS made it slower than ext2, but now with soft updates, that performance difference is gone so you have two files systems, ext2 and ufs, similar peformance, but one is crash-safe and the other is not. And, when comparing the journalling file systems, you have UFS vs. XFS/ext3/JFS/Reiser, and UFS is faster. The only thing the journalling file system give you is more rapid reboot, but frankly, if your OS goes down often enough so that is an issue, you have bigger problems than fsync time. The big problem is that Linux went from non-crash safe right to crash-safe and reboot quick. We need a middle ground, which is where UFS/soft updates is. > The fact that ext2 defaults to asynchronous mode and UFS (at least on > the BSDs) defaults to synchronous mode seems like a total non-issue to > me. Is there any more to the alleged difference in reliability? The reliability problem isn't alleged. ext2 developers admits ext2 isn't 100% crash-safe. They will say it is usually crash-safe, but that isn't good enough for PostgreSQL. I wish I was wrong. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073 From pgsql-general-owner@postgresql.org Thu Sep 26 17:42:05 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 969E9476FEF; Thu, 26 Sep 2002 17:42:02 -0400 (EDT) Received: from candle.pha.pa.us (momjian.navpoint.com [207.106.42.251]) by postgresql.org (Postfix) with ESMTP id 97422476FDB; Thu, 26 Sep 2002 17:42:00 -0400 (EDT) Received: (from pgman@localhost) by candle.pha.pa.us (8.11.6/8.10.1) id g8QLfMr09064; Thu, 26 Sep 2002 17:41:22 -0400 (EDT) From: Bruce Momjian Message-Id: <200209262141.g8QLfMr09064@candle.pha.pa.us> Subject: Re: [HACKERS] Performance while loading data and indexing In-Reply-To: To: Doug McNaught Date: Thu, 26 Sep 2002 17:41:22 -0400 (EDT) Cc: Tom Lane , Neil Conway , shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org X-Mailer: ELM [version 2.4ME+ PL99 (25)] MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1225 X-Sequence-Number: 30806 Doug McNaught wrote: > Tom Lane writes: > > > We'd be happiest with a filesystem that journals its own metadata and > > not the user data in the file(s). I dunno if there are any. > > ext3 with data=writeback? (See my previous message to Bruce). OK, so that makes ext3 crash safe without lots of overhead? -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073 From pgsql-hackers-owner@postgresql.org Thu Sep 26 17:45:16 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 460CF477047; Thu, 26 Sep 2002 17:45:14 -0400 (EDT) Received: from squire.barchord.com (squire.barchord.com [216.194.67.18]) by postgresql.org (Postfix) with ESMTP id CA66D477042; Thu, 26 Sep 2002 17:45:13 -0400 (EDT) Received: from [192.168.1.253] (CPE00508b028d7d-CM00803785c5e0.cpe.net.cable.rogers.com [24.103.51.175]) by squire.barchord.com (Postfix) with ESMTP id 454C542C; Thu, 26 Sep 2002 17:45:15 -0400 (EDT) Subject: Re: [GENERAL] Performance while loading data and indexing From: Rod Taylor To: Bruce Momjian Cc: Neil Conway , Greg Copeland , shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List In-Reply-To: <200209262139.g8QLdEE08861@candle.pha.pa.us> References: <200209262139.g8QLdEE08861@candle.pha.pa.us> Content-Type: text/plain Content-Transfer-Encoding: 7bit X-Mailer: Ximian Evolution 1.0.8 Date: 26 Sep 2002 17:45:23 -0400 Message-Id: <1033076723.27772.4.camel@jester> Mime-Version: 1.0 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1598 X-Sequence-Number: 29555 On Thu, 2002-09-26 at 17:39, Bruce Momjian wrote: > Neil Conway wrote: > > Greg Copeland writes: > > > On Thu, 2002-09-26 at 16:03, Neil Conway wrote: > > > > I'm not really familiar with the reasoning behind ext2's > > > > reputation as recovering poorly from crashes; if we fsync a WAL > > > > record to disk before we lose power, can't we recover reliably, > > > > even with ext2? > > > > > > Well, I have experienced data loss from ext2 before. Also, recovery > > > from crashes on large file systems take a very, very long time. > > > > Yes, but wouldn't you face exactly the same issues if you ran a > > UFS-like filesystem in asynchronous mode? Albeit it's not the default, > > but performance in synchronous mode is usually pretty poor. > > Yes, before UFS had soft updates, the synchronous nature of UFS made it > slower than ext2, but now with soft updates, that performance difference > is gone so you have two files systems, ext2 and ufs, similar peformance, > but one is crash-safe and the other is not. Note entirely true. ufs is both crash-safe and quick-rebootable. You do need to fsck at some point, but not prior to mounting it. Any corrupt blocks are empty, and are easy to avoid. Someone just needs to implement a background fsck that will run on a mounted filesystem. -- Rod Taylor From pgsql-general-owner@postgresql.org Thu Sep 26 17:48:03 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 42A774767DA; Thu, 26 Sep 2002 17:48:03 -0400 (EDT) Received: from candle.pha.pa.us (momjian.navpoint.com [207.106.42.251]) by postgresql.org (Postfix) with ESMTP id 5FC1E476212; Thu, 26 Sep 2002 17:48:01 -0400 (EDT) Received: (from pgman@localhost) by candle.pha.pa.us (8.11.6/8.10.1) id g8QLlhU10159; Thu, 26 Sep 2002 17:47:43 -0400 (EDT) From: Bruce Momjian Message-Id: <200209262147.g8QLlhU10159@candle.pha.pa.us> Subject: Re: [HACKERS] Performance while loading data and indexing In-Reply-To: <1033076723.27772.4.camel@jester> To: Rod Taylor Date: Thu, 26 Sep 2002 17:47:43 -0400 (EDT) Cc: Neil Conway , Greg Copeland , shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List X-Mailer: ELM [version 2.4ME+ PL99 (25)] MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1227 X-Sequence-Number: 30808 Rod Taylor wrote: > > Yes, before UFS had soft updates, the synchronous nature of UFS made it > > slower than ext2, but now with soft updates, that performance difference > > is gone so you have two files systems, ext2 and ufs, similar peformance, > > but one is crash-safe and the other is not. > > Note entirely true. ufs is both crash-safe and quick-rebootable. You > do need to fsck at some point, but not prior to mounting it. Any > corrupt blocks are empty, and are easy to avoid. I am assuming you need to mount the drive as part of the reboot. Of course you can boot fast with any file system if you don't have to mount it. :-) -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073 From pgsql-general-owner@postgresql.org Sat Sep 28 13:41:23 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 991FA47618E; Thu, 26 Sep 2002 18:03:27 -0400 (EDT) Received: from squire.barchord.com (squire.barchord.com [216.194.67.18]) by postgresql.org (Postfix) with ESMTP id 221B74760B7; Thu, 26 Sep 2002 18:03:27 -0400 (EDT) Received: from [192.168.1.253] (CPE00508b028d7d-CM00803785c5e0.cpe.net.cable.rogers.com [24.103.51.175]) by squire.barchord.com (Postfix) with ESMTP id 8ECED42C; Thu, 26 Sep 2002 18:03:28 -0400 (EDT) Subject: Re: [HACKERS] Performance while loading data and indexing From: Rod Taylor To: Bruce Momjian Cc: Neil Conway , Greg Copeland , shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List In-Reply-To: <200209262147.g8QLlhU10159@candle.pha.pa.us> References: <200209262147.g8QLlhU10159@candle.pha.pa.us> Content-Type: text/plain Content-Transfer-Encoding: 7bit X-Mailer: Ximian Evolution 1.0.8 Date: 26 Sep 2002 18:03:36 -0400 Message-Id: <1033077816.27772.9.camel@jester> Mime-Version: 1.0 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1330 X-Sequence-Number: 30911 On Thu, 2002-09-26 at 17:47, Bruce Momjian wrote: > Rod Taylor wrote: > > > Yes, before UFS had soft updates, the synchronous nature of UFS made it > > > slower than ext2, but now with soft updates, that performance difference > > > is gone so you have two files systems, ext2 and ufs, similar peformance, > > > but one is crash-safe and the other is not. > > > > Note entirely true. ufs is both crash-safe and quick-rebootable. You > > do need to fsck at some point, but not prior to mounting it. Any > > corrupt blocks are empty, and are easy to avoid. > > I am assuming you need to mount the drive as part of the reboot. Of > course you can boot fast with any file system if you don't have to mount > it. :-) Sorry, poor explanation. Background fsck (when implemented) would operate on a currently mounted (and active) file system. The only reason fsck is required prior to reboot now is because no-one had done the work. http://www.freebsd.org/cgi/man.cgi?query=fsck&sektion=8&manpath=FreeBSD+5.0-current See the first paragraph of the above. -- Rod Taylor From pgsql-general-owner@postgresql.org Thu Sep 26 18:05:13 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id C0E5F476FAF; Thu, 26 Sep 2002 18:05:12 -0400 (EDT) Received: from candle.pha.pa.us (momjian.navpoint.com [207.106.42.251]) by postgresql.org (Postfix) with ESMTP id C71D6476F95; Thu, 26 Sep 2002 18:05:10 -0400 (EDT) Received: (from pgman@localhost) by candle.pha.pa.us (8.11.6/8.10.1) id g8QM4qX11641; Thu, 26 Sep 2002 18:04:52 -0400 (EDT) From: Bruce Momjian Message-Id: <200209262204.g8QM4qX11641@candle.pha.pa.us> Subject: Re: [HACKERS] Performance while loading data and indexing In-Reply-To: <1033077816.27772.9.camel@jester> To: Rod Taylor Date: Thu, 26 Sep 2002 18:04:52 -0400 (EDT) Cc: Neil Conway , Greg Copeland , shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List X-Mailer: ELM [version 2.4ME+ PL99 (25)] MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1229 X-Sequence-Number: 30810 Rod Taylor wrote: > On Thu, 2002-09-26 at 17:47, Bruce Momjian wrote: > > Rod Taylor wrote: > > > > Yes, before UFS had soft updates, the synchronous nature of UFS made it > > > > slower than ext2, but now with soft updates, that performance difference > > > > is gone so you have two files systems, ext2 and ufs, similar peformance, > > > > but one is crash-safe and the other is not. > > > > > > Note entirely true. ufs is both crash-safe and quick-rebootable. You > > > do need to fsck at some point, but not prior to mounting it. Any > > > corrupt blocks are empty, and are easy to avoid. > > > > I am assuming you need to mount the drive as part of the reboot. Of > > course you can boot fast with any file system if you don't have to mount > > it. :-) > > Sorry, poor explanation. > > Background fsck (when implemented) would operate on a currently mounted > (and active) file system. The only reason fsck is required prior to > reboot now is because no-one had done the work. > > http://www.freebsd.org/cgi/man.cgi?query=fsck&sektion=8&manpath=FreeBSD+5.0-current > > See the first paragraph of the above. Oh, yes, I have heard of that missing feature. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073 From pgsql-general-owner@postgresql.org Thu Sep 26 19:26:22 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 625C74763DD; Thu, 26 Sep 2002 19:26:21 -0400 (EDT) Received: from varsoon.wireboard.com (www.wireboard.com [216.151.155.101]) by postgresql.org (Postfix) with ESMTP id C8EF147628D; Thu, 26 Sep 2002 19:26:20 -0400 (EDT) Received: from doug by varsoon.wireboard.com with local (Exim 3.35 #1) id 17ui1D-0006sX-00; Thu, 26 Sep 2002 19:26:03 -0400 To: Bruce Momjian From: Doug McNaught Cc: Tom Lane , Neil Conway , shridhar_daithankar@persistent.co.in, pgsql-hackers@postgresql.org, pgsql-general@postgresql.org Subject: Re: [HACKERS] Performance while loading data and indexing References: <200209262141.g8QLfMr09064@candle.pha.pa.us> Date: 26 Sep 2002 19:26:03 -0400 In-Reply-To: Bruce Momjian's message of "Thu, 26 Sep 2002 17:41:22 -0400 (EDT)" Message-ID: Lines: 23 User-Agent: Gnus/5.0806 (Gnus v5.8.6) Emacs/20.7 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1231 X-Sequence-Number: 30812 Bruce Momjian writes: > Doug McNaught wrote: > > Tom Lane writes: > > > > > We'd be happiest with a filesystem that journals its own metadata and > > > not the user data in the file(s). I dunno if there are any. > > > > ext3 with data=writeback? (See my previous message to Bruce). > > OK, so that makes ext3 crash safe without lots of overhead? Metadata is journaled so you shouldn't lose data blocks or directory entries. Some data blocks (that haven't been fsync()'ed) may have old or wrong data in them, but I think that's the same as ufs, right? And WAL replay should take care of that. It'd be very interesting to do some tests of the various journaling modes. I have an old K6 that I might be able to turn into a hit-the-reset-switch-at-ramdom-times machine. What kind of tests should be run? -Doug From pgsql-performance-owner@postgresql.org Thu Sep 26 22:53:21 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 353A4476391 for ; Thu, 26 Sep 2002 22:53:21 -0400 (EDT) Received: from corpmail.outblaze.com (202-77-223-51.outblaze.com [202.77.223.51]) by postgresql.org (Postfix) with ESMTP id D37A247632D for ; Thu, 26 Sep 2002 22:53:19 -0400 (EDT) Received: from yusufg.portal2.com (202-77-223-2.outblaze.com [202.77.223.2]) by corpmail.outblaze.com (8.12.6/8.12.6) with SMTP id g8R2rMm8029328 for ; Fri, 27 Sep 2002 02:53:22 GMT Received: (qmail 1463 invoked by uid 500); 27 Sep 2002 02:55:10 -0000 Date: Fri, 27 Sep 2002 10:55:10 +0800 From: Yusuf Goolamabbas To: pgsql-performance@postgresql.org Subject: Would ext3 data=journal help for Postgres synchronous io mode Message-ID: <20020927025510.GB1175@outblaze.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.4i X-AntiVirus: checked by Vexira Milter 1.0.0.3; VAE 6.15.0.1; VDF 6.15.0.9 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/14 X-Sequence-Number: 14 According to ext3 hackers (Stephen Tweedie, Andrew Morton). ext3 data=journal mode is much faster than any of the other mode for workloads which do a lot of syncrhonous i/o. Personally, I have seen dramatic improvements on moving mail queues to this mode (postfix in particularly flies with this mode) While this may seem contradictory (forcing journaling for the data in addition to the metadata), it will likely improve the performance for sync I/O loads like mail servers because it can do all of the I/O to the journal without any seek or sync overhead while the mail is arriving. I assume that since Postgresql does a lot of fsyncs, it would benefit also. I have sent email to Sridhar asking if he could test this Another thing to note is that Linux 2.4.x kernels < 2.4.20-pre4 use bounce buffer's to do IO if the machine has > 1GB memory. Distributor kernels such as Redhat/Suse/Mandrake are patched to do IO via DMA to/from highmem (>1GB). According to IBM's paper @ OLS, this improves IO performance by 40% BTW, Is this list archived on the website Regards, Yusuf -- Yusuf Goolamabbas yusufg@outblaze.com From pgsql-general-owner@postgresql.org Thu Sep 26 23:08:32 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 3C676477068 for ; Thu, 26 Sep 2002 23:08:30 -0400 (EDT) Received: from sss.pgh.pa.us (unknown [192.204.191.242]) by postgresql.org (Postfix) with ESMTP id 3674347703B for ; Thu, 26 Sep 2002 23:08:29 -0400 (EDT) Received: from sss2.sss.pgh.pa.us (tgl@localhost [127.0.0.1]) by sss.pgh.pa.us (8.12.5/8.12.5) with ESMTP id g8R37jhR020360; Thu, 26 Sep 2002 23:07:45 -0400 (EDT) To: Doug McNaught Cc: Bruce Momjian , Greg Copeland , shridhar_daithankar@persistent.co.in, PostgresSQL Hackers Mailing List , PostgresSQL General Mailing List Subject: Re: [HACKERS] Performance while loading data and indexing In-reply-to: References: <200209262000.g8QK0mG10553@candle.pha.pa.us> Comments: In-reply-to Doug McNaught message dated "26 Sep 2002 17:31:55 -0400" Date: Thu, 26 Sep 2002 23:07:44 -0400 Message-ID: <20359.1033096064@sss.pgh.pa.us> From: Tom Lane X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1242 X-Sequence-Number: 30823 Doug McNaught writes: > "data=writeback" means that no data is journaled, just metadata (which > is like XFS or Reiser). An fsync() call should still do what it > normally does, commit the writes to disk before returning. > "data=journal" journals all data and is the slowest and safest. > "data=ordered" writes out data blocks before committing a journal > transaction, which is faster than full data journaling (since data > doesn't get written twice) and almost as safe. "data=writeback" is > noted to keep obsolete data in the case of some crashes (since the > data may not have been written yet) but a completed fsync() should > ensure that the data is valid. Thanks for the explanation. > So I guess I'd probably use data=ordered for an all-on-one-fs > installation, and data=writeback for a WAL-only drive. Actually I think the ideal thing for Postgres would be data=writeback for both data and WAL drives. We can handle loss of un-fsync'd data for ourselves in both cases. Of course, if you keep anything besides Postgres data files on a partition, you'd possibly want the more secure settings. regards, tom lane From pgsql-general-owner@postgresql.org Fri Sep 27 01:12:31 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id D61BE47714E; Fri, 27 Sep 2002 01:12:30 -0400 (EDT) Received: from panda.center-f1.ru (panda.center-f1.ru [195.151.30.15]) by postgresql.org (Postfix) with ESMTP id C4927477196; Fri, 27 Sep 2002 01:12:28 -0400 (EDT) Received: from localhost (byg@localhost) by panda.center-f1.ru (8.11.6/8.11.6) with ESMTP id g8R5Ee308324; Fri, 27 Sep 2002 12:14:40 +0700 X-Authentication-Warning: panda.center-f1.ru: byg owned process doing -bs Date: Fri, 27 Sep 2002 12:14:40 +0700 (NOVST) From: Yury Bokhoncovich To: Bruce Momjian Cc: Neil Conway , , , Subject: Re: [HACKERS] Performance while loading data and indexing In-Reply-To: <200209262107.g8QL7vN25965@candle.pha.pa.us> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1253 X-Sequence-Number: 30834 Hello! On Thu, 26 Sep 2002, Bruce Momjian wrote: > > I'm not really familiar with the reasoning behind ext2's reputation as > > recovering poorly from crashes; if we fsync a WAL record to disk On relatively big volumes ext2 recovery can end up in formatting the fs under certain cirrumstances.;-) > > > I assumed it was the double fsync for the normal and journal that > > > made the journalling file systems slog. > > > > Well, a journalling file system would need to write a journal entry > > and flush that to disk, even if fsync is disabled -- whereas without > > fsync enabled, ext2 doesn't have to flush anything to disk. ISTM that > > the performance advantage of ext2 over ext3 is should be even larger > > when fsync is not enabled. > > Yes, it is still double-writing. I just thought that if that wasn't > happening while the db was waiting for a commit that it wouldn't be too > bad. > > Is it just me or do all the Linux file systems seem like they are > lacking something when PostgreSQL is concerned? We just want a UFS-like > file system on Linux and no one has it. mount -o sync an ext2 volume on Linux - and you can get a "UFS-like" fs.:) mount -o async an FFS volume on FreeBSD - and you can get boost in fs performance. Personally me always mount ext2 fs where Pg is living with sync option. Fsync in pg is off (since 6.3), this way successfully pass thru a few serious crashes on various systems (mostly on power problems). If fsync is on in Pg, performance gets so-oh-oh-oh-oh slowly!=) I just have done upgrade from 2.2 kernel on ext2 to ext3 capable 2.4 one so I'm planning to do some benchmarking. Roughly saying w/o benchmarks, the performance have been degraded in 2/3 proportion. "But better safe then sorry". -- WBR, Yury Bokhoncovich, Senior System Administrator, NOC of F1 Group. Phone: +7 (3832) 106228, ext.140, E-mail: byg@center-f1.ru. Unix is like a wigwam -- no Gates, no Windows, and an Apache inside. From pgsql-performance-owner@postgresql.org Fri Sep 27 05:42:31 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 95321475EDF for ; Fri, 27 Sep 2002 05:42:26 -0400 (EDT) Received: from freemail.agrinet.ch (freemail.agrinet.ch [212.28.134.90]) by postgresql.org (Postfix) with ESMTP id 267654771AE for ; Fri, 27 Sep 2002 05:42:23 -0400 (EDT) Received: from syydelaervli.fortytwo.ch (81.6.8.94) by freemail.agrinet.ch (NPlex 5.1.056) id 3D921E0A00007388 for pgsql-performance@postgresql.org; Fri, 27 Sep 2002 11:42:24 +0200 Received: from atlas.acter.ch (unknown [212.126.160.108]) by syydelaervli.fortytwo.ch (Postfix) with ESMTP id 43EF52A76 for ; Fri, 27 Sep 2002 11:42:23 +0200 (CEST) Received: by atlas.acter.ch (Postfix, from userid 1047) id 7B2A89696; Fri, 27 Sep 2002 11:42:22 +0200 (CEST) Subject: From: Adrian von Bidder To: PostgreSQL Performance Mailing List Content-Type: text/plain Content-Transfer-Encoding: 7bit X-Mailer: Ximian Evolution 1.0.8 Date: 27 Sep 2002 11:42:22 +0200 Message-Id: <1033119742.13843.9.camel@atlas> Mime-Version: 1.0 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/15 X-Sequence-Number: 15 subscribe -- secure email with gpg http://fortytwo.ch/gpg NOTICE: subkey signature! request key 92082481 from keyserver.kjsl.com From pgsql-general-owner@postgresql.org Fri Sep 27 06:40:18 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 883AA475D00 for ; Fri, 27 Sep 2002 06:40:17 -0400 (EDT) Received: from news.hub.org (news.hub.org [64.49.215.80]) by postgresql.org (Postfix) with ESMTP id 3C44B475EC7 for ; Fri, 27 Sep 2002 06:40:16 -0400 (EDT) Received: by news.hub.org (Postfix, from userid 8) id 5AC2C381683; Fri, 27 Sep 2002 06:40:15 -0400 (EDT) From: Mats Lofkvist X-Newsgroups: comp.databases.postgresql.general, comp.databases.postgresql.questions Subject: Re: [HACKERS] Performance while loading data and indexing Date: 27 Sep 2002 12:40:13 +0200 Organization: Hub.Org Networking Services (http://www.hub.org) Lines: 29 Message-ID: References: <87n0q4zc0l.fsf@mailbox.samurai.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Complaints-To: news@hub.org User-Agent: Gnus/5.0803 (Gnus v5.8.3) Emacs/20.5 To: pgsql-general@postgresql.org X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1268 X-Sequence-Number: 30849 neilc@samurai.com (Neil Conway) writes: [snip] > > Well, I have experienced data loss from ext2 before. Also, recovery > > from crashes on large file systems take a very, very long time. > > Yes, but wouldn't you face exactly the same issues if you ran a > UFS-like filesystem in asynchronous mode? Albeit it's not the default, > but performance in synchronous mode is usually pretty poor. > > The fact that ext2 defaults to asynchronous mode and UFS (at least on > the BSDs) defaults to synchronous mode seems like a total non-issue to > me. Is there any more to the alleged difference in reliability? UFS on most unix systems (BSD, solaris etc) defaults to sync metadata, async data which is a mode that is completely missing from ext2 as far as I know. This is why UFS is considered safer than ext2. (Running with 'sync' is too slow to be a usable alternative in most cases.) _ Mats Lofkvist mal@algonet.se PS The BSD soft updates yields the safety of the default sync metadata / async data mode while being at least as fast as running fully async. From pgsql-general-owner@postgresql.org Fri Sep 27 06:49:23 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id CD832476243 for ; Fri, 27 Sep 2002 06:49:21 -0400 (EDT) Received: from news.hub.org (news.hub.org [64.49.215.80]) by postgresql.org (Postfix) with ESMTP id 69480476160 for ; Fri, 27 Sep 2002 06:49:21 -0400 (EDT) Received: by news.hub.org (Postfix, from userid 8) id 68B29381683; Fri, 27 Sep 2002 06:49:18 -0400 (EDT) From: Mats Lofkvist X-Newsgroups: comp.databases.postgresql.general, comp.databases.postgresql.questions Subject: Re: Performance while loading data and indexing Date: 27 Sep 2002 12:49:17 +0200 Organization: Hub.Org Networking Services (http://www.hub.org) Lines: 21 Message-ID: References: <3D931438.22010.133ADAFA@localhost> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Complaints-To: news@hub.org User-Agent: Gnus/5.0803 (Gnus v5.8.3) Emacs/20.5 To: pgsql-general@postgresql.org X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1269 X-Sequence-Number: 30850 shridhar_daithankar@persistent.co.in ("Shridhar Daithankar") writes: [snip] > > Couple MB of data per sec. to disk is just not saturating it. It's a RAID 5 > setup.. > RAID5 is not the best for performance, especially write performance. If it is software RAID it is even worse :-). (Note also that you need to check that you are not saturating the number of seeks the disks can handle, not just the bandwith.) Striping should be better (combined with mirroring if you need the safety, but with both striping and mirroring you may need multiple SCSI channels). _ Mats Lofkvist mal@algonet.se From pgsql-general-owner@postgresql.org Fri Sep 27 11:20:38 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 99F9947608D for ; Fri, 27 Sep 2002 11:20:36 -0400 (EDT) Received: from mail1.ihs.com (mail1.ihs.com [170.207.70.222]) by postgresql.org (Postfix) with ESMTP id BAD85475FEB for ; Fri, 27 Sep 2002 11:20:35 -0400 (EDT) Received: from css120.ihs.com (css120.ihs.com [170.207.105.120]) by mail1.ihs.com (8.12.6/8.12.6) with ESMTP id g8RFJikY015704; Fri, 27 Sep 2002 09:19:45 -0600 (MDT) Date: Fri, 27 Sep 2002 09:16:03 -0600 (MDT) From: "scott.marlowe" To: Mats Lofkvist Cc: Subject: Re: Performance while loading data and indexing In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-MailBodyFilter: Message body has not been filtered X-MailScanner: Found to be clean X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1280 X-Sequence-Number: 30861 On 27 Sep 2002, Mats Lofkvist wrote: > shridhar_daithankar@persistent.co.in ("Shridhar Daithankar") writes: > > [snip] > > > > Couple MB of data per sec. to disk is just not saturating it. It's a RAID 5 > > setup.. > > > > RAID5 is not the best for performance, especially write performance. > If it is software RAID it is even worse :-). I take exception to this. RAID5 is a great choice for most folks. 1: RAID5 only writes out the parity stripe and data stripe, not all stripes when writing. So, in an 8 disk RAID5 array, writing to a single 64 k stripe involves one 64k read (parity stripe) and two 64k writes. On a mirror set, writing to one 64k stripe involves two 64k writes. The difference isn't that great, and in my testing, a large enough RAID5 provides so much faster read speads by spreading the reads across so many heads as to more than make up for the slightly slower writes. My testing has shown that a 4 disk RAID5 can generally run about 85% or more the speed of a mirror set. 2: Why does EVERYONE have to jump on the bandwagon that software RAID 5 is bad. My workstation running RH 7.2 uses about 1% of the CPU during very heavy parallel access (i.e. 50 simo pgbenchs) at most. I've seen many hardware RAID cards that are noticeable slower than my workstation running software RAID. You do know that hardware RAID is just software RAID where the processing is done on a seperate CPU on a card, but it's still software doing the work. 3: We just had a hardware RAID card mark both drives in a mirror set bad. It wouldn't accept them back, and all the data was gone. poof. That would never happen in Linux's kernel software RAID, I can always make Linux take back a "bad" drive. The only difference between RAID5 with n+1 disks and RAID0 with n disks is that we have to write a parity stripe in RAID5. It's ability to handle high parallel load is much better than a RAID1 set, and on average, you actually write about the same amount with either RAID1 or RAID5. Don't dog software RAID5, it works and it works well in Linux. Windows, however, is another issue. There, the software RAID5 is pretty pitiful, both in terms of performance and maintenance. From pgsql-hackers-owner@postgresql.org Fri Sep 27 15:01:43 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 976E6476D74 for ; Fri, 27 Sep 2002 15:01:42 -0400 (EDT) Received: from Mail.CERT.Uni-Stuttgart.DE (mail.cert.uni-stuttgart.de [129.69.16.17]) by postgresql.org (Postfix) with ESMTP id 3F3234761C2 for ; Fri, 27 Sep 2002 15:01:38 -0400 (EDT) Received: from rusfw by Mail.CERT.Uni-Stuttgart.DE with local (Exim 4.04) id 17v0Ms-0006Sx-00 for pgsql-hackers@postgresql.org; Fri, 27 Sep 2002 21:01:38 +0200 To: pgsql-hackers@postgresql.org Subject: Re: [GENERAL] Performance while loading data and indexing References: <200209262045.g8QKjtv21744@candle.pha.pa.us> <87vg4szco1.fsf@mailbox.samurai.com> <12930.1033075921@sss.pgh.pa.us> Mail-Followup-To: pgsql-hackers@postgresql.org From: Florian Weimer Date: Fri, 27 Sep 2002 21:01:38 +0200 In-Reply-To: <12930.1033075921@sss.pgh.pa.us> (Tom Lane's message of "Thu, 26 Sep 2002 17:32:01 -0400") Message-ID: <87hegbjlyl.fsf@Login.CERT.Uni-Stuttgart.DE> Lines: 12 User-Agent: Gnus/5.090007 (Oort Gnus v0.07) Emacs/21.2 (i386-debian-linux-gnu) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1630 X-Sequence-Number: 29587 Tom Lane writes: > We'd be happiest with a filesystem that journals its own metadata and > not the user data in the file(s). I dunno if there are any. Most journalling file systems work this way. Data journalling is not very widespread, AFAIK. -- Florian Weimer Weimer@CERT.Uni-Stuttgart.DE University of Stuttgart http://CERT.Uni-Stuttgart.DE/people/fw/ RUS-CERT fax +49-711-685-5898 From pgsql-general-owner@postgresql.org Fri Sep 27 21:46:57 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id CBFF34762B5 for ; Fri, 27 Sep 2002 21:46:55 -0400 (EDT) Received: from deborah.paradise.net.nz (deborah.paradise.net.nz [203.96.152.32]) by postgresql.org (Postfix) with ESMTP id 107AB47626D for ; Fri, 27 Sep 2002 21:46:55 -0400 (EDT) Received: from paradise.net.nz (203-79-74-52.adsl.paradise.net.nz [203.79.74.52]) by deborah.paradise.net.nz (Postfix) with ESMTP id 73EE1D1CAF; Sat, 28 Sep 2002 13:46:48 +1200 (NZST) Message-ID: <3D95082C.4070904@paradise.net.nz> Date: Sat, 28 Sep 2002 13:38:52 +1200 From: Mark Kirkwood User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:0.9.8) Gecko/20020204 X-Accept-Language: en-us MIME-Version: 1.0 To: "scott.marlowe" Cc: Mats Lofkvist , pgsql-general@postgresql.org Subject: Re: Performance while loading data and indexing References: Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/1319 X-Sequence-Number: 30900 scott.marlowe wrote: >(snippage) >I take exception to this. RAID5 is a great choice for most folks. > > I agree - certainly RAID5 *used* to be rather sad, but modern cards have improved this no end on the hardware side - e.g. I recently benchmarked a 3Ware 8x card on a system with 4 x 15000 rpm Maxtor 70Gb drives and achieved 120 Mb/s for (8K) reads and 60 Mb/s for (8K) writes using RAID5. I used Redhat 7.3 + ext2. The benchmarking program was Bonnie. Given that the performance of a single disk was ~30 Mb/s for reads and writes, I felt this was quite a good result ! ( Other cards I had tried previously struggled to maintain 1/2 the write rate of a single disk in such a configuration). As for software RAID5, I have not tried it out. Of course I could not get 60Mb/s while COPYing data into Postgres... typically cpu seemed to be the bottleneck in this case (what was the actual write rate? I hear you asking..err.. cant recall I'm afraid.. must try it out again ) cheers Mark From pgsql-performance-owner@postgresql.org Sat Sep 28 11:49:21 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 7544B475E26 for ; Sat, 28 Sep 2002 11:49:20 -0400 (EDT) Received: from kumudu.nslk.com (kumudu.nslk.com [64.247.55.254]) by postgresql.org (Postfix) with ESMTP id 01B30475E25 for ; Sat, 28 Sep 2002 11:49:20 -0400 (EDT) Received: from [203.94.93.204] (helo=W) by kumudu.nslk.com with asmtp (Exim 3.36 #1) id 17vJqE-00034v-00 for pgsql-performance@postgresql.org; Sat, 28 Sep 2002 11:49:15 -0400 Message-ID: <010801c26706$be9e3820$cc5d5ecb@A.GEEKIYANAGE> Reply-To: "Waruna Geekiyanage" From: "Waruna Geekiyanage" To: Subject: INDEX Date: Sat, 28 Sep 2002 21:50:13 +0600 MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="----=_NextPart_000_0105_01C26739.06687120" X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 5.00.2919.6600 X-MimeOLE: Produced By Microsoft MimeOLE V5.00.2919.6600 X-AntiAbuse: This header was added to track abuse, please include it with any abuse report X-AntiAbuse: Primary Hostname - kumudu.nslk.com X-AntiAbuse: Original Domain - postgresql.org X-AntiAbuse: Originator/Caller UID/GID - [0 0] / [0 0] X-AntiAbuse: Sender Address Domain - nirmani.com X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/16 X-Sequence-Number: 16 This is a multi-part message in MIME format. ------=_NextPart_000_0105_01C26739.06687120 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable When a table is created with a primary key it generates a index. Dos the queries on that table use that index automatically? Do I need to reindex that index after insertions? ------=_NextPart_000_0105_01C26739.06687120 Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable
When a table is created with a primary key= it=20 generates a index.
Dos the queries on that table use that ind= ex=20 automatically?
Do I need to reindex that index after=20 insertions?
------=_NextPart_000_0105_01C26739.06687120-- From pgsql-performance-owner@postgresql.org Sat Sep 28 15:13:19 2002 Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id 5E9C84762E3 for ; Sat, 28 Sep 2002 15:13:18 -0400 (EDT) Received: from mail.libertyrms.com (unknown [209.167.124.227]) by postgresql.org (Postfix) with ESMTP id B81B647612B for ; Sat, 28 Sep 2002 15:13:17 -0400 (EDT) Received: from andrew by mail.libertyrms.com with local (Exim 3.22 #3 (Debian)) id 17vN1i-00065x-00 for ; Sat, 28 Sep 2002 15:13:18 -0400 Date: Sat, 28 Sep 2002 15:13:18 -0400 From: Andrew Sullivan To: pgsql-performance@postgresql.org Subject: Re: INDEX Message-ID: <20020928151318.B22793@mail.libertyrms.com> Mail-Followup-To: pgsql-performance@postgresql.org References: <010801c26706$be9e3820$cc5d5ecb@A.GEEKIYANAGE> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.2.5i In-Reply-To: <010801c26706$be9e3820$cc5d5ecb@A.GEEKIYANAGE>; from waruna@nirmani.com on Sat, Sep 28, 2002 at 09:50:13PM +0600 X-Virus-Scanned: by AMaViS new-20020517 X-Archive-Number: 200209/17 X-Sequence-Number: 17 On Sat, Sep 28, 2002 at 09:50:13PM +0600, Waruna Geekiyanage wrote: > When a table is created with a primary key it generates a index. > Dos the queries on that table use that index automatically? Only if you analyse the table, and it's a "win". See the various past discussion on -general, for instance, about index use, and the FAQ. > Do I need to reindex that index after insertions? No, but you need to analyse. A -- ---- Andrew Sullivan 204-4141 Yonge Street Liberty RMS Toronto, Ontario Canada M2P 2A8 +1 416 646 3304 x110