From owner-pgsql-hackers@hub.org Sun Jun 14 18:45:04 1998 Received: from hub.org (hub.org [209.47.148.200]) by candle.pha.pa.us (8.8.5/8.8.5) with ESMTP id SAA03690 for ; Sun, 14 Jun 1998 18:45:00 -0400 (EDT) Received: from localhost (majordom@localhost) by hub.org (8.8.8/8.7.5) with SMTP id SAA28049; Sun, 14 Jun 1998 18:39:42 -0400 (EDT) Received: by hub.org (TLB v0.10a (1.23 tibbs 1997/01/09 00:29:32)); Sun, 14 Jun 1998 18:36:06 +0000 (EDT) Received: (from majordom@localhost) by hub.org (8.8.8/8.7.5) id SAA27943 for pgsql-hackers-outgoing; Sun, 14 Jun 1998 18:36:04 -0400 (EDT) Received: from angular.illustra.com (ifmxoak.illustra.com [206.175.10.34]) by hub.org (8.8.8/8.7.5) with ESMTP id SAA27925 for ; Sun, 14 Jun 1998 18:35:47 -0400 (EDT) Received: from hawk.illustra.com (hawk.illustra.com [158.58.61.70]) by angular.illustra.com (8.7.4/8.7.3) with SMTP id PAA21293 for ; Sun, 14 Jun 1998 15:35:12 -0700 (PDT) Received: by hawk.illustra.com (5.x/smail2.5/06-10-94/S) id AA07922; Sun, 14 Jun 1998 15:35:13 -0700 From: dg@illustra.com (David Gould) Message-Id: <9806142235.AA07922@hawk.illustra.com> Subject: [HACKERS] performance tests, initial results To: pgsql-hackers@postgreSQL.org Date: Sun, 14 Jun 1998 15:35:13 -0700 (PDT) Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-pgsql-hackers@hub.org Precedence: bulk Status: RO I have been playing a little with the performance tests found in pgsql/src/tests/performance and have a few observations that might be of minor interest. The tests themselves are simple enough although the result parsing in the driver did not work on Linux. I am enclosing a patch below to fix this. I think it will also work better on the other systems. A summary of results from my testing are below. Details are at the bottom of this message. My test system is 'leslie': linux 2.0.32, gcc version 2.7.2.3 P133, HX chipset, 512K L2, 32MB mem NCR810 fast scsi, Quantum Atlas 2GB drive (7200 rpm). Results Summary (times in seconds) Single txn 8K txn Create 8K idx 8K random Simple Case Description 8K insert 8K insert Index Insert Scans Orderby =================== ========== ========= ====== ====== ========= ======= 1 From Distribution P90 FreeBsd -B256 39.56 1190.98 3.69 46.65 65.49 2.27 IDE 2 Running on leslie P133 Linux 2.0.32 15.48 326.75 2.99 20.69 35.81 1.68 SCSI 32M 3 leslie, -o -F no forced writes 15.90 24.98 2.63 20.46 36.43 1.69 4 leslie, -o -F no ASSERTS 14.92 23.23 1.38 18.67 33.79 1.58 5 leslie, -o -F -B2048 more buffers 21.31 42.28 2.65 25.74 42.26 1.72 6 leslie, -o -F -B2048 more bufs, no ASSERT 20.52 39.79 1.40 24.77 39.51 1.55 Case to Case Difference Factors (+ is faster) Single txn 8K txn Create 8K idx 8K random Simple Case Description 8K insert 8K insert Index Insert Scans Orderby =================== ========== ========= ====== ====== ========= ======= leslie vs BSD P90. 2.56 3.65 1.23 2.25 1.83 1.35 (noflush -F) vs no -F -1.03 13.08 1.14 1.01 -1.02 1.00 No Assert vs Assert 1.05 1.07 1.90 1.06 1.07 1.09 -B256 vs -B2048 1.34 1.69 1.01 1.26 1.16 1.02 Observations: - leslie (P133 linux) appears to be about 1.8 times faster than the P90 BSD system used for the test result distributed with the source, not counting the 8K txn insert case which was completely disk bound. - SCSI disks make a big (factor of 3.6) difference. During this test the disk was hammering and cpu utilization was < 10%. - Assertion checking seems to cost about 7% except for create index where it costs 90% - the -F option to avoid flushing buffers has tremendous effect if there are many very small transactions. Or, another way, flushing at the end of the transaction is a major disaster for performance. - Something is very wrong with our buffer cache implementation. Going from 256 buffers to 2048 buffers costs an average of 25%. In the 8K txn case it costs about 70%. I see looking at the code and profiling that in the 8K txn case this is in BufferSync() which examines all the buffers at commit time. I don't quite understand why it is so costly for the single 8K row txn (35%) though. It would be nice to have some more tests. Maybe the Wisconsin stuff will be useful. ----------------- patch to test harness. apply from pgsql ------------ *** src/test/performance/runtests.pl.orig Sun Jun 14 11:34:04 1998 Differences % ----------------- patch to test harness. apply from pgsql ------------ *** src/test/performance/runtests.pl.orig Sun Jun 14 11:34:04 1998 --- src/test/performance/runtests.pl Sun Jun 14 12:07:30 1998 *************** *** 84,123 **** open (STDERR, ">$TmpFile") or die; select (STDERR); $| = 1; ! for ($i = 0; $i <= $#perftests; $i++) ! { $test = $perftests[$i]; ($test, $XACTBLOCK) = split (/ /, $test); $runtest = $test; ! if ( $test =~ /\.ntm/ ) ! { ! # # No timing for this queries - # close (STDERR); # close $TmpFile open (STDERR, ">/dev/null") or die; $runtest =~ s/\.ntm//; } ! else ! { close (STDOUT); open(STDOUT, ">&SAVEOUT"); print STDOUT "\nRunning: $perftests[$i+1] ..."; close (STDOUT); open (STDOUT, ">/dev/null") or die; select (STDERR); $| = 1; ! printf "$perftests[$i+1]: "; } do "sqls/$runtest"; # Restore STDERR to $TmpFile ! if ( $test =~ /\.ntm/ ) ! { close (STDERR); open (STDERR, ">>$TmpFile") or die; } - select (STDERR); $| = 1; $i++; } --- 84,116 ---- open (STDERR, ">$TmpFile") or die; select (STDERR); $| = 1; ! for ($i = 0; $i <= $#perftests; $i++) { $test = $perftests[$i]; ($test, $XACTBLOCK) = split (/ /, $test); $runtest = $test; ! if ( $test =~ /\.ntm/ ) { # No timing for this queries close (STDERR); # close $TmpFile open (STDERR, ">/dev/null") or die; $runtest =~ s/\.ntm//; } ! else { close (STDOUT); open(STDOUT, ">&SAVEOUT"); print STDOUT "\nRunning: $perftests[$i+1] ..."; close (STDOUT); open (STDOUT, ">/dev/null") or die; select (STDERR); $| = 1; ! print "$perftests[$i+1]: "; } do "sqls/$runtest"; # Restore STDERR to $TmpFile ! if ( $test =~ /\.ntm/ ) { close (STDERR); open (STDERR, ">>$TmpFile") or die; } select (STDERR); $| = 1; $i++; } *************** *** 128,138 **** open (TMPF, "<$TmpFile") or die; open (RESF, ">$ResFile") or die; ! while () ! { ! $str = $_; ! ($test, $rtime) = split (/:/, $str); ! ($tmp, $rtime, $rest) = split (/[ ]+/, $rtime); ! print RESF "$test: $rtime\n"; } --- 121,130 ---- open (TMPF, "<$TmpFile") or die; open (RESF, ">$ResFile") or die; ! while () { ! if (m/^(.*: ).* ([0-9:.]+) *elapsed/) { ! ($test, $rtime) = ($1, $2); ! print RESF $test, $rtime, "\n"; ! } } ------------------------------------------------------------------------ ------------------------- testcase detail -------------------------- 1. from distribution DBMS: PostgreSQL 6.2b10 OS: FreeBSD 2.1.5-RELEASE HardWare: i586/90, 24M RAM, IDE StartUp: postmaster -B 256 '-o -S 2048' -S Compiler: gcc 2.6.3 Compiled: -O, without CASSERT checking, with -DTBL_FREE_CMD_MEMORY (to free memory if BEGIN/END after each query execution) DB connection startup: 0.20 8192 INSERTs INTO SIMPLE (1 xact): 39.58 8192 INSERTs INTO SIMPLE (8192 xacts): 1190.98 Create INDEX on SIMPLE: 3.69 8192 INSERTs INTO SIMPLE with INDEX (1 xact): 46.65 8192 random INDEX scans on SIMPLE (1 xact): 65.49 ORDER BY SIMPLE: 2.27 2. run on leslie with asserts DBMS: PostgreSQL 6.3.2 (plus changes to 98/06/01) OS: Linux 2.0.32 leslie HardWare: i586/133 HX 512, 32M RAM, fast SCSI, 7200rpm StartUp: postmaster -B 256 '-o -S 2048' -S Compiler: gcc 2.7.2.3 Compiled: -O, WITH CASSERT checking, with -DTBL_FREE_CMD_MEMORY (to free memory if BEGIN/END after each query execution) DB connection startup: 0.10 8192 INSERTs INTO SIMPLE (1 xact): 15.48 8192 INSERTs INTO SIMPLE (8192 xacts): 326.75 Create INDEX on SIMPLE: 2.99 8192 INSERTs INTO SIMPLE with INDEX (1 xact): 20.69 8192 random INDEX scans on SIMPLE (1 xact): 35.81 ORDER BY SIMPLE: 1.68 3. with -F to avoid forced i/o DBMS: PostgreSQL 6.3.2 (plus changes to 98/06/01) OS: Linux 2.0.32 leslie HardWare: i586/133 HX 512, 32M RAM, fast SCSI, 7200rpm StartUp: postmaster -B 256 '-o -S 2048 -F' -S Compiler: gcc 2.7.2.3 Compiled: -O, WITH CASSERT checking, with -DTBL_FREE_CMD_MEMORY (to free memory if BEGIN/END after each query execution) DB connection startup: 0.10 8192 INSERTs INTO SIMPLE (1 xact): 15.90 8192 INSERTs INTO SIMPLE (8192 xacts): 24.98 Create INDEX on SIMPLE: 2.63 8192 INSERTs INTO SIMPLE with INDEX (1 xact): 20.46 8192 random INDEX scans on SIMPLE (1 xact): 36.43 ORDER BY SIMPLE: 1.69 4. no asserts, -F to avoid forced I/O DBMS: PostgreSQL 6.3.2 (plus changes to 98/06/01) OS: Linux 2.0.32 leslie HardWare: i586/133 HX 512, 32M RAM, fast SCSI, 7200rpm StartUp: postmaster -B 256 '-o -S 2048' -S Compiler: gcc 2.7.2.3 Compiled: -O, No CASSERT checking, with -DTBL_FREE_CMD_MEMORY (to free memory if BEGIN/END after each query execution) DB connection startup: 0.10 8192 INSERTs INTO SIMPLE (1 xact): 14.92 8192 INSERTs INTO SIMPLE (8192 xacts): 23.23 Create INDEX on SIMPLE: 1.38 8192 INSERTs INTO SIMPLE with INDEX (1 xact): 18.67 8192 random INDEX scans on SIMPLE (1 xact): 33.79 ORDER BY SIMPLE: 1.58 5. with more buffers (2048 vs 256) and -F to avoid forced i/o DBMS: PostgreSQL 6.3.2 (plus changes to 98/06/01) OS: Linux 2.0.32 leslie HardWare: i586/133 HX 512, 32M RAM, fast SCSI, 7200rpm StartUp: postmaster -B 2048 '-o -S 2048 -F' -S Compiler: gcc 2.7.2.3 Compiled: -O, WITH CASSERT checking, with -DTBL_FREE_CMD_MEMORY (to free memory if BEGIN/END after each query execution) DB connection startup: 0.11 8192 INSERTs INTO SIMPLE (1 xact): 21.31 8192 INSERTs INTO SIMPLE (8192 xacts): 42.28 Create INDEX on SIMPLE: 2.65 8192 INSERTs INTO SIMPLE with INDEX (1 xact): 25.74 8192 random INDEX scans on SIMPLE (1 xact): 42.26 ORDER BY SIMPLE: 1.72 6. No Asserts, more buffers (2048 vs 256) and -F to avoid forced i/o DBMS: PostgreSQL 6.3.2 (plus changes to 98/06/01) OS: Linux 2.0.32 leslie HardWare: i586/133 HX 512, 32M RAM, fast SCSI, 7200rpm StartUp: postmaster -B 2048 '-o -S 2048 -F' -S Compiler: gcc 2.7.2.3 Compiled: -O, No CASSERT checking, with -DTBL_FREE_CMD_MEMORY (to free memory if BEGIN/END after each query execution) DB connection startup: 0.11 8192 INSERTs INTO SIMPLE (1 xact): 20.52 8192 INSERTs INTO SIMPLE (8192 xacts): 39.79 Create INDEX on SIMPLE: 1.40 8192 INSERTs INTO SIMPLE with INDEX (1 xact): 24.77 8192 random INDEX scans on SIMPLE (1 xact): 39.51 ORDER BY SIMPLE: 1.55 --------------------------------------------------------------------- -dg David Gould dg@illustra.com 510.628.3783 or 510.305.9468 Informix Software (No, really) 300 Lakeside Drive Oakland, CA 94612 "Don't worry about people stealing your ideas. If your ideas are any good, you'll have to ram them down people's throats." -- Howard Aiken From owner-pgsql-hackers@hub.org Tue Oct 19 10:31:10 1999 Received: from renoir.op.net (root@renoir.op.net [209.152.193.4]) by candle.pha.pa.us (8.9.0/8.9.0) with ESMTP id KAA29087 for ; Tue, 19 Oct 1999 10:31:08 -0400 (EDT) Received: from hub.org (hub.org [216.126.84.1]) by renoir.op.net (o1/$Revision: 1.16 $) with ESMTP id KAA27535 for ; Tue, 19 Oct 1999 10:19:47 -0400 (EDT) Received: from localhost (majordom@localhost) by hub.org (8.9.3/8.9.3) with SMTP id KAA30328; Tue, 19 Oct 1999 10:12:10 -0400 (EDT) (envelope-from owner-pgsql-hackers) Received: by hub.org (bulk_mailer v1.5); Tue, 19 Oct 1999 10:11:55 -0400 Received: (from majordom@localhost) by hub.org (8.9.3/8.9.3) id KAA30030 for pgsql-hackers-outgoing; Tue, 19 Oct 1999 10:11:00 -0400 (EDT) (envelope-from owner-pgsql-hackers@postgreSQL.org) Received: from sss.sss.pgh.pa.us (sss.pgh.pa.us [209.114.166.2]) by hub.org (8.9.3/8.9.3) with ESMTP id KAA29914 for ; Tue, 19 Oct 1999 10:10:33 -0400 (EDT) (envelope-from tgl@sss.pgh.pa.us) Received: from sss.sss.pgh.pa.us (localhost [127.0.0.1]) by sss.sss.pgh.pa.us (8.9.1/8.9.1) with ESMTP id KAA09038; Tue, 19 Oct 1999 10:09:15 -0400 (EDT) To: "Hiroshi Inoue" cc: "Vadim Mikheev" , pgsql-hackers@postgreSQL.org Subject: Re: [HACKERS] mdnblocks is an amazing time sink in huge relations In-reply-to: Your message of Tue, 19 Oct 1999 19:03:22 +0900 <000801bf1a19$2d88ae20$2801007e@cadzone.tpf.co.jp> Date: Tue, 19 Oct 1999 10:09:15 -0400 Message-ID: <9036.940342155@sss.pgh.pa.us> From: Tom Lane Sender: owner-pgsql-hackers@postgreSQL.org Status: RO "Hiroshi Inoue" writes: > 1. shared cache holds committed system tuples. > 2. private cache holds uncommitted system tuples. > 3. relpages of shared cache are updated immediately by > phisical change and corresponding buffer pages are > marked dirty. > 4. on commit, the contents of uncommitted tuples except > relpages,reltuples,... are copied to correponding tuples > in shared cache and the combined contents are > committed. > If so,catalog cache invalidation would be no longer needed. > But synchronization of the step 4. may be difficult. I think the main problem is that relpages and reltuples shouldn't be kept in pg_class columns at all, because they need to have very different update behavior from the other pg_class columns. The rest of pg_class is update-on-commit, and we can lock down any one row in the normal MVCC way (if transaction A has modified a row and transaction B also wants to modify it, B waits for A to commit or abort, so it can know which version of the row to start from). Furthermore, there can legitimately be several different values of a row in use in different places: the latest committed, an uncommitted modification, and one or more old values that are still being used by active transactions because they were current when those transactions started. (BTW, the present relcache is pretty bad about maintaining pure MVCC transaction semantics like this, but it seems clear to me that that's the direction we want to go in.) relpages cannot operate this way. To be useful for avoiding lseeks, relpages *must* change exactly when the physical file changes. It matters not at all whether the particular transaction that extended the file ultimately commits or not. Moreover there can be only one correct value (per relation) across the whole system, because there is only one length of the relation file. If we want to take reltuples seriously and try to maintain it on-the-fly, then I think it needs still a third behavior. Clearly it cannot be updated using MVCC rules, or we lose all writer concurrency (if A has added tuples to a rel, B would have to wait for A to commit before it could update reltuples...). Furthermore "updating" isn't a simple matter of storing what you think the new value is; otherwise two transactions adding tuples in parallel would leave the wrong answer after B commits and overwrites A's value. I think it would work for each transaction to keep track of a net delta in reltuples for each table it's changed (total tuples added less total tuples deleted), and then atomically add that value to the table's shared reltuples counter during commit. But that still leaves the problem of how you use the counter during a transaction to get an accurate answer to the question "If I scan this table now, how many tuples will I see?" At the time the question is asked, the current shared counter value might include the effects of transactions that have committed since your transaction started, and therefore are not visible under MVCC rules. I think getting the correct answer would involve making an instantaneous copy of the current counter at the start of your xact, and then adding your own private net-uncommitted-delta to the saved shared counter value when asked the question. This doesn't look real practical --- you'd have to save the reltuples counts of *all* tables in the database at the start of each xact, on the off chance that you might need them. Ugh. Perhaps someone has a better idea. In any case, reltuples clearly needs different mechanisms than the ordinary fields in pg_class do, because updating it will be a performance bottleneck otherwise. If we allow reltuples to be updated only by vacuum-like events, as it is now, then I think keeping it in pg_class is still OK. In short, it seems clear to me that relpages should be removed from pg_class and kept somewhere else if we want to make it more reliable than it is now, and the same for reltuples (but reltuples doesn't behave the same as relpages, and probably ought to be handled differently). regards, tom lane ************ From owner-pgsql-hackers@hub.org Tue Oct 19 21:25:30 1999 Received: from renoir.op.net (root@renoir.op.net [209.152.193.4]) by candle.pha.pa.us (8.9.0/8.9.0) with ESMTP id VAA28130 for ; Tue, 19 Oct 1999 21:25:26 -0400 (EDT) Received: from hub.org (hub.org [216.126.84.1]) by renoir.op.net (o1/$Revision: 1.16 $) with ESMTP id VAA10512 for ; Tue, 19 Oct 1999 21:15:28 -0400 (EDT) Received: from localhost (majordom@localhost) by hub.org (8.9.3/8.9.3) with SMTP id VAA50745; Tue, 19 Oct 1999 21:07:23 -0400 (EDT) (envelope-from owner-pgsql-hackers) Received: by hub.org (bulk_mailer v1.5); Tue, 19 Oct 1999 21:07:01 -0400 Received: (from majordom@localhost) by hub.org (8.9.3/8.9.3) id VAA50644 for pgsql-hackers-outgoing; Tue, 19 Oct 1999 21:06:06 -0400 (EDT) (envelope-from owner-pgsql-hackers@postgreSQL.org) Received: from sd.tpf.co.jp (sd.tpf.co.jp [210.161.239.34]) by hub.org (8.9.3/8.9.3) with ESMTP id VAA50584 for ; Tue, 19 Oct 1999 21:05:26 -0400 (EDT) (envelope-from Inoue@tpf.co.jp) Received: from cadzone ([126.0.1.40] (may be forged)) by sd.tpf.co.jp (2.5 Build 2640 (Berkeley 8.8.6)/8.8.4) with SMTP id KAA01715; Wed, 20 Oct 1999 10:05:14 +0900 From: "Hiroshi Inoue" To: "Tom Lane" Cc: Subject: RE: [HACKERS] mdnblocks is an amazing time sink in huge relations Date: Wed, 20 Oct 1999 10:09:13 +0900 Message-ID: <000501bf1a97$b925a860$2801007e@cadzone.tpf.co.jp> 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 8.5, Build 4.71.2173.0 X-Mimeole: Produced By Microsoft MimeOLE V4.72.2106.4 Importance: Normal Sender: owner-pgsql-hackers@postgreSQL.org Status: RO > -----Original Message----- > From: Hiroshi Inoue [mailto:Inoue@tpf.co.jp] > Sent: Tuesday, October 19, 1999 6:45 PM > To: Tom Lane > Cc: pgsql-hackers@postgreSQL.org > Subject: RE: [HACKERS] mdnblocks is an amazing time sink in huge > relations > > > > > > "Hiroshi Inoue" writes: > > [snip] > > > > > > Deletion is necessary only not to consume disk space. > > > > > > For example vacuum could remove not deleted files. > > > > Hmm ... interesting idea ... but I can hear the complaints > > from users already... > > > > My idea is only an analogy of PostgreSQL's simple recovery > mechanism of tuples. > > And my main point is > "delete fails after commit" doesn't harm the database > except that not deleted files consume disk space. > > Of cource,it's preferable to delete relation files immediately > after(or just when) commit. > Useless files are visible though useless tuples are invisible. > Anyway I don't need "DROP TABLE inside transactions" now and my idea is originally for that issue. After a thought,I propose the following solution. 1. mdcreate() couldn't create existent relation files. If the existent file is of length zero,we would overwrite the file.(seems the comment in md.c says so but the code doesn't do so). If the file is an Index relation file,we would overwrite the file. 2. mdunlink() couldn't unlink non-existent relation files. mdunlink() doesn't call elog(ERROR) even if the file doesn't exist,though I couldn't find where to change now. mdopen() doesn't call elog(ERROR) even if the file doesn't exist and leaves the relation as CLOSED. Comments ? Regards. Hiroshi Inoue Inoue@tpf.co.jp ************ From pgsql-hackers-owner+M6267@hub.org Sun Aug 27 21:46:37 2000 Received: from hub.org (root@hub.org [216.126.84.1]) by candle.pha.pa.us (8.9.0/8.9.0) with ESMTP id UAA07972 for ; Sun, 27 Aug 2000 20:46:36 -0400 (EDT) Received: from hub.org (majordom@localhost [127.0.0.1]) by hub.org (8.10.1/8.10.1) with SMTP id e7S0kaL27996; Sun, 27 Aug 2000 20:46:36 -0400 (EDT) Received: from sss2.sss.pgh.pa.us (sss.pgh.pa.us [209.114.166.2]) by hub.org (8.10.1/8.10.1) with ESMTP id e7S05aL24107 for ; Sun, 27 Aug 2000 20:05:36 -0400 (EDT) Received: from sss2.sss.pgh.pa.us (tgl@localhost [127.0.0.1]) by sss2.sss.pgh.pa.us (8.9.3/8.9.3) with ESMTP id UAA01604 for ; Sun, 27 Aug 2000 20:05:29 -0400 (EDT) To: pgsql-hackers@postgreSQL.org Subject: [HACKERS] Possible performance improvement: buffer replacement policy Date: Sun, 27 Aug 2000 20:05:29 -0400 Message-ID: <1601.967421129@sss.pgh.pa.us> From: Tom Lane X-Mailing-List: pgsql-hackers@postgresql.org Precedence: bulk Sender: pgsql-hackers-owner@hub.org Status: RO Those of you with long memories may recall a benchmark that Edmund Mergl drew our attention to back in May '99. That test showed extremely slow performance for updating a table with many indexes (about 20). At the time, it seemed the problem was due to bad performance of btree with many equal keys, so I thought I'd go back and retry the benchmark after this latest round of btree hackery. The good news is that btree itself seems to be pretty well fixed; the bad news is that the benchmark is still slow for large numbers of rows. The problem is I/O: the CPU mostly sits idle waiting for the disk. As best I can tell, the difficulty is that the working set of pages needed to update this many indexes is too large compared to the number of disk buffers Postgres is using. (I was running with -B 1000 and looking at behavior for a 100000-row test table. This gave me a table size of 3876 pages, plus 11526 pages in 20 indexes.) Of course, there's only so much we can do when the number of buffers is too small, but I still started to wonder if we are using the buffers as effectively as we can. Some tracing showed that most of the pages of the indexes were being read and written multiple times within a single UPDATE query, while most of the pages of the table proper were fetched and written only once. That says we're not using the buffers as well as we could; the index pages are not being kept in memory when they should be. In a query like this, we should displace main-table pages sooner to allow keeping more index pages in cache --- but with the simple LRU replacement method we use, once a page has been loaded it will stay in cache for at least the next NBuffers (-B) page references, no matter what. With a large NBuffers that's a long time. I've come across an interesting article: The LRU-K Page Replacement Algorithm For Database Disk Buffering Elizabeth J. O'Neil, Patrick E. O'Neil, Gerhard Weikum Proceedings of the 1993 ACM SIGMOD international conference on Management of Data, May 1993 (If you subscribe to the ACM digital library, you can get a PDF of this from there.) This article argues that standard LRU buffer management is inherently not great for database caches, and that it's much better to replace pages on the basis of time since the K'th most recent reference, not just time since the most recent one. K=2 is enough to get most of the benefit. The big win is that you are measuring an actual page interreference time (between the last two references) and not just dealing with a lower-bound guess on the interreference time. Frequently used pages are thus much more likely to stay in cache. It looks like it wouldn't take too much work to replace shared buffers on the basis of LRU-2 instead of LRU, so I'm thinking about trying it. Has anyone looked into this area? Is there a better method to try? regards, tom lane From prlw1@newn.cam.ac.uk Fri Jan 19 12:54:45 2001 Received: from henry.newn.cam.ac.uk (henry.newn.cam.ac.uk [131.111.204.130]) by candle.pha.pa.us (8.9.0/8.9.0) with ESMTP id MAA29822 for ; Fri, 19 Jan 2001 12:54:44 -0500 (EST) Received: from [131.111.204.180] (helo=quartz.newn.cam.ac.uk) by henry.newn.cam.ac.uk with esmtp (Exim 3.13 #1) id 14JfkU-0001WA-00; Fri, 19 Jan 2001 17:54:54 +0000 Received: from prlw1 by quartz.newn.cam.ac.uk with local (Exim 3.13 #1) id 14Jfj6-0001cq-00; Fri, 19 Jan 2001 17:53:28 +0000 Date: Fri, 19 Jan 2001 17:53:28 +0000 From: Patrick Welche To: Bruce Momjian Cc: Tom Lane , pgsql-hackers@postgreSQL.org Subject: Re: [HACKERS] Possible performance improvement: buffer replacement policy Message-ID: <20010119175328.A6223@quartz.newn.cam.ac.uk> Reply-To: prlw1@cam.ac.uk References: <1601.967421129@sss.pgh.pa.us> <200101191703.MAA25873@candle.pha.pa.us> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.2i In-Reply-To: <200101191703.MAA25873@candle.pha.pa.us>; from pgman@candle.pha.pa.us on Fri, Jan 19, 2001 at 12:03:58PM -0500 Status: RO On Fri, Jan 19, 2001 at 12:03:58PM -0500, Bruce Momjian wrote: > > Tom, did we ever test this? I think we did and found that it was the > same or worse, right? (Funnily enough, I just read that message:) To: Bruce Momjian cc: pgsql-hackers@postgreSQL.org Subject: Re: [HACKERS] Possible performance improvement: buffer replacement policy In-reply-to: <200010161541.LAA06653@candle.pha.pa.us> References: <200010161541.LAA06653@candle.pha.pa.us> Comments: In-reply-to Bruce Momjian message dated "Mon, 16 Oct 2000 11:41:41 -0400" Date: Mon, 16 Oct 2000 11:49:52 -0400 Message-ID: <26100.971711392@sss.pgh.pa.us> From: Tom Lane X-Mailing-List: pgsql-hackers@postgresql.org Precedence: bulk Sender: pgsql-hackers-owner@hub.org Status: RO Content-Length: 947 Lines: 19 Bruce Momjian writes: >> It looks like it wouldn't take too much work to replace shared buffers >> on the basis of LRU-2 instead of LRU, so I'm thinking about trying it. >> >> Has anyone looked into this area? Is there a better method to try? > Sounds like a perfect idea. Good luck. :-) Actually, the idea went down in flames :-(, but I neglected to report back to pghackers about it. I did do some code to manage buffers as LRU-2. I didn't have any good performance test cases to try it with, but Richard Brosnahan was kind enough to re-run the TPC tests previously published by Great Bridge with that code in place. Wasn't any faster, in fact possibly a little slower, likely due to the extra CPU time spent on buffer freelist management. It's possible that other scenarios might show a better result, but right now I feel pretty discouraged about the LRU-2 idea and am not pursuing it. regards, tom lane From pgsql-hackers-owner+M3455@postgresql.org Fri Jan 19 13:18:12 2001 Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by candle.pha.pa.us (8.9.0/8.9.0) with ESMTP id NAA02092 for ; Fri, 19 Jan 2001 13:18:12 -0500 (EST) Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by mail.postgresql.org (8.11.1/8.11.1) with SMTP id f0JIFJ037872; Fri, 19 Jan 2001 13:15:19 -0500 (EST) (envelope-from pgsql-hackers-owner+M3455@postgresql.org) Received: from sectorbase2.sectorbase.com ([208.48.122.131]) by mail.postgresql.org (8.11.1/8.11.1) with ESMTP id f0JI7V036780 for ; Fri, 19 Jan 2001 13:07:31 -0500 (EST) (envelope-from vmikheev@SECTORBASE.COM) Received: by sectorbase2.sectorbase.com with Internet Mail Service (5.5.2653.19) id ; Fri, 19 Jan 2001 09:46:14 -0800 Message-ID: <8F4C99C66D04D4118F580090272A7A234D329F@sectorbase1.sectorbase.com> From: "Mikheev, Vadim" To: "'Tom Lane'" , Bruce Momjian Cc: pgsql-hackers@postgresql.org Subject: RE: [HACKERS] Possible performance improvement: buffer replacemen t policy Date: Fri, 19 Jan 2001 10:07:27 -0800 MIME-Version: 1.0 X-Mailer: Internet Mail Service (5.5.2653.19) Content-Type: text/plain; charset="iso-8859-1" Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org Status: RO > > Tom, did we ever test this? I think we did and found that > > it was the same or worse, right? > > I tried it and didn't see any noticeable improvement on the particular > test case I was using, so I got discouraged and didn't pursue the idea > further. I'd like to come back to it someday, though. I don't know how much useful could be LRU-2 but with WAL we should try to reuse undirty free buffers first, not dirty ones, just to postpone writes as long as we can. (BTW, this is what Oracle does.) So, we probably should put new unfree dirty buffer just before first dirty one in LRU. Vadim From markw@mohawksoft.com Thu Jun 7 14:40:02 2001 Return-path: Received: from gromit.dotclick.com (ipn9-f8366.net-resource.net [216.204.83.66]) by candle.pha.pa.us (8.10.1/8.10.1) with ESMTP id f57Ie1c14004 for ; Thu, 7 Jun 2001 14:40:02 -0400 (EDT) Received: from mohawksoft.com (IDENT:markw@localhost.localdomain [127.0.0.1]) by gromit.dotclick.com (8.9.3/8.9.3) with ESMTP id OAA04973; Thu, 7 Jun 2001 14:37:00 -0400 Sender: markw@gromit.dotclick.com Message-ID: <3B1FC9CB.57C72AD6@mohawksoft.com> Date: Thu, 07 Jun 2001 14:36:59 -0400 From: mlw X-Mailer: Mozilla 4.75 [en] (X11; U; Linux 2.4.2 i686) X-Accept-Language: en MIME-Version: 1.0 To: Bruce Momjian , "pgsql-hackers@postgresql.org" Subject: Re: 7.2 items References: <200106071503.f57F32n03924@candle.pha.pa.us> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Status: RO Bruce Momjian wrote: > > Bruce Momjian writes: > > > > > Here is a small list of big TODO items. I was wondering which ones > > > people were thinking about for 7.2? > > > > A friend of mine wants to use PostgreSQL instead of Oracle for a large > > application, but has run into a snag when speed comparisons looked > > good until the Oracle folks added a couple of BITMAP indexes. I can't > > recall seeing any discussion about that here -- are there any plans? > > It is not on our list and I am not sure what they do. Do you have access to any Oracle Documentation? There is a good explanation of them. However, I will try to explain. If you have a table, locations. It has 1,000,000 records. In oracle you do this: create bitmap index bitmap_foo on locations (state) ; For each unique value of 'state' oracle will create a bitmap with 1,000,000 bits in it. With a one representing a match and a zero representing no match. Record '0' in the table is represented by bit '0' in the bitmap, record '1' is represented by bit '1', record two by bit '2' and so on. In a table where comparatively few different values are to be indexed in a large table, a bitmap index can be quite small and not suffer the N * log(N) disk I/O most tree based indexes suffer. If the bitmap is fairly sparse or dense (or have periods of denseness and sparseness), it can be compressed very efficiently as well. When the statement: select * from locations where state = 'MA'; Is executed, the bitmap is read into memory in very few disk operations. (Perhaps even as few as one or two). It is a simple operation of rifling through the bitmap for '1's that indicate the record has the property, 'state' = 'MA'; From mascarm@mascari.com Thu Jun 7 15:36:25 2001 Return-path: Received: from corvette.mascari.com (dhcp065-024-161-045.columbus.rr.com [65.24.161.45]) by candle.pha.pa.us (8.10.1/8.10.1) with ESMTP id f57JaOc21943 for ; Thu, 7 Jun 2001 15:36:24 -0400 (EDT) Received: from ferrari (ferrari.mascari.com [192.168.2.1]) by corvette.mascari.com (8.9.3/8.9.3) with SMTP id PAA25607; Thu, 7 Jun 2001 15:29:31 -0400 Received: by localhost with Microsoft MAPI; Thu, 7 Jun 2001 15:34:18 -0400 Message-ID: <01C0EF67.5105D2E0.mascarm@mascari.com> From: Mike Mascari Reply-To: "mascarm@mascari.com" To: "'mlw'" , Bruce Momjian , "pgsql-hackers@postgresql.org" Subject: RE: [HACKERS] Re: 7.2 items Date: Thu, 7 Jun 2001 15:34:17 -0400 Organization: Mascari Development Inc. X-Mailer: Microsoft Internet E-mail/MAPI - 8.0.0.4211 MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Status: RO And in addition, If you submitted the query: SELECT * FROM addresses WHERE state = 'OH' AND areacode = '614' Then, with bitmap indexes, the bitmaps are just logically ANDed together, and the final bitmap determines the matching rows. Mike Mascari mascarm@mascari.com -----Original Message----- From: mlw [SMTP:markw@mohawksoft.com] Bruce Momjian wrote: > > Bruce Momjian writes: > > > > > Here is a small list of big TODO items. I was wondering which ones > > > people were thinking about for 7.2? > > > > A friend of mine wants to use PostgreSQL instead of Oracle for a large > > application, but has run into a snag when speed comparisons looked > > good until the Oracle folks added a couple of BITMAP indexes. I can't > > recall seeing any discussion about that here -- are there any plans? > > It is not on our list and I am not sure what they do. Do you have access to any Oracle Documentation? There is a good explanation of them. However, I will try to explain. If you have a table, locations. It has 1,000,000 records. In oracle you do this: create bitmap index bitmap_foo on locations (state) ; For each unique value of 'state' oracle will create a bitmap with 1,000,000 bits in it. With a one representing a match and a zero representing no match. Record '0' in the table is represented by bit '0' in the bitmap, record '1' is represented by bit '1', record two by bit '2' and so on. In a table where comparatively few different values are to be indexed in a large table, a bitmap index can be quite small and not suffer the N * log(N) disk I/O most tree based indexes suffer. If the bitmap is fairly sparse or dense (or have periods of denseness and sparseness), it can be compressed very efficiently as well. When the statement: select * from locations where state = 'MA'; Is executed, the bitmap is read into memory in very few disk operations. (Perhaps even as few as one or two). It is a simple operation of rifling through the bitmap for '1's that indicate the record has the property, 'state' = 'MA'; From oleg@sai.msu.su Thu Jun 7 15:39:15 2001 Return-path: Received: from ra.sai.msu.su (ra.sai.msu.su [158.250.29.2]) by candle.pha.pa.us (8.10.1/8.10.1) with ESMTP id f57Jd7c22010 for ; Thu, 7 Jun 2001 15:39:08 -0400 (EDT) Received: from ra (ra [158.250.29.2]) by ra.sai.msu.su (8.9.3/8.9.3) with ESMTP id WAA07783; Thu, 7 Jun 2001 22:38:20 +0300 (GMT) Date: Thu, 7 Jun 2001 22:38:20 +0300 (GMT) From: Oleg Bartunov X-X-Sender: To: mlw cc: Bruce Momjian , "pgsql-hackers@postgresql.org" Subject: Re: [HACKERS] Re: 7.2 items In-Reply-To: <3B1FC9CB.57C72AD6@mohawksoft.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Status: RO I think it's possible to implement bitmap indexes with a little effort using GiST. at least I know one implementation http://www.it.iitb.ernet.in/~rvijay/dbms/proj/ if you have interests you could implement bitmap indexes yourself unfortunately, we're very busy Oleg On Thu, 7 Jun 2001, mlw wrote: > Bruce Momjian wrote: > > > > Bruce Momjian writes: > > > > > > > Here is a small list of big TODO items. I was wondering which ones > > > > people were thinking about for 7.2? > > > > > > A friend of mine wants to use PostgreSQL instead of Oracle for a large > > > application, but has run into a snag when speed comparisons looked > > > good until the Oracle folks added a couple of BITMAP indexes. I can't > > > recall seeing any discussion about that here -- are there any plans? > > > > It is not on our list and I am not sure what they do. > > Do you have access to any Oracle Documentation? There is a good explanation > of them. > > However, I will try to explain. > > If you have a table, locations. It has 1,000,000 records. > > In oracle you do this: > > create bitmap index bitmap_foo on locations (state) ; > > For each unique value of 'state' oracle will create a bitmap with 1,000,000 > bits in it. With a one representing a match and a zero representing no > match. Record '0' in the table is represented by bit '0' in the bitmap, > record '1' is represented by bit '1', record two by bit '2' and so on. > > In a table where comparatively few different values are to be indexed in a > large table, a bitmap index can be quite small and not suffer the N * log(N) > disk I/O most tree based indexes suffer. If the bitmap is fairly sparse or > dense (or have periods of denseness and sparseness), it can be compressed > very efficiently as well. > > When the statement: > > select * from locations where state = 'MA'; > > Is executed, the bitmap is read into memory in very few disk operations. > (Perhaps even as few as one or two). It is a simple operation of rifling > through the bitmap for '1's that indicate the record has the property, > 'state' = 'MA'; > > > ---------------------------(end of broadcast)--------------------------- > TIP 6: Have you searched our list archives? > > http://www.postgresql.org/search.mpl > Regards, Oleg _____________________________________________________________ Oleg Bartunov, sci.researcher, hostmaster of AstroNet, Sternberg Astronomical Institute, Moscow University (Russia) Internet: oleg@sai.msu.su, http://www.sai.msu.su/~megera/ phone: +007(095)939-16-83, +007(095)939-23-83 From pgsql-general-owner+M2497@hub.org Fri Jun 16 18:31:03 2000 Received: from renoir.op.net (root@renoir.op.net [207.29.195.4]) by candle.pha.pa.us (8.9.0/8.9.0) with ESMTP id RAA04165 for ; Fri, 16 Jun 2000 17:31:01 -0400 (EDT) Received: from hub.org (root@hub.org [216.126.84.1]) by renoir.op.net (o1/$Revision: 1.16 $) with ESMTP id RAA13110 for ; Fri, 16 Jun 2000 17:20:12 -0400 (EDT) Received: from hub.org (majordom@localhost [127.0.0.1]) by hub.org (8.10.1/8.10.1) with SMTP id e5GLDaM14477; Fri, 16 Jun 2000 17:13:36 -0400 (EDT) Received: from home.dialix.com ([203.15.150.26]) by hub.org (8.10.1/8.10.1) with ESMTP id e5GLCQM14064 for ; Fri, 16 Jun 2000 17:12:27 -0400 (EDT) Received: from nemeton.com.au ([202.76.153.71]) by home.dialix.com (8.9.3/8.9.3/JustNet) with SMTP id HAA95516 for ; Sat, 17 Jun 2000 07:11:44 +1000 (EST) (envelope-from giles@nemeton.com.au) Received: (qmail 10213 invoked from network); 16 Jun 2000 09:52:29 -0000 Received: from nemeton.com.au (203.8.3.17) by nemeton.com.au with SMTP; 16 Jun 2000 09:52:29 -0000 To: Jurgen Defurne cc: Mark Stier , postgreSQL general mailing list Subject: Re: [GENERAL] optimization by removing the file system layer? In-Reply-To: Message from Jurgen Defurne of "Thu, 15 Jun 2000 20:26:57 +0200." <39491FF1.E1E583F8@glo.be> Date: Fri, 16 Jun 2000 19:52:28 +1000 Message-ID: <10210.961149148@nemeton.com.au> From: Giles Lean X-Mailing-List: pgsql-general@postgresql.org Precedence: bulk Sender: pgsql-general-owner@hub.org Status: OR > I think that the Un*x filesystem is one of the reasons that large > database vendors rather use raw devices, than filesystem storage > files. This used to be the preference, back in the late 80s and possibly early 90s. I'm seeing a preference toward using the filesystem now, possibly with some sort of async I/O and co-operation from the OS filesystem about interactions with the filesystem cache. Performance preferences don't stand still. The hardware changes, the software changes, the volume of data changes, and different solutions become preferable. > Using a raw device on the disk gives them the possibility to have > complete control over their files, indices and objects without being > bothered by the operating system. > > This speeds up things in several ways : > - the least possible OS intervention Not that this is especially useful, necessarily. If the "raw" device is in fact managed by a logical volume manager doing mirroring onto some sort of storage array there is still plenty of OS code involved. The cost of using a filesystem in addition may not be much if anything and of course a filesystem is considerably more flexible to administer (backup, move, change size, check integrity, etc.) > - choose block sizes according to applications > - reducing fragmentation > - packing data in nearby cilinders ... but when this storage area is spread over multiple mechanisms in a smart storage array with write caching, you've no idea what is where anyway. Better to let the hardware or at least the OS manage this; there are so many levels of caching between a database and the magnetic media that working hard to influence layout is almost certainly a waste of time. Kirk McKusick tells a lovely story that once upon a time it used to be sensible to check some registers on a particular disk controller to find out where the heads were when scheduling I/O. Needless to say, that is history now! There's a considerable cost in complexity and code in using "raw" storage too, and it's not a one off cost: as the technologies change, the "fast" way to do things will change and the code will have to be updated to match. Better to leave this to the OS vendor where possible, and take advantage of the tuning they do. > - Anyone other ideas -> the sky is the limit here > It also aids portability, at least on platforms that have an > equivalent of a raw device. I don't understand that claim. Not much is portable about raw devices, and they're typically not nearlly as well documented as the filesystem interfaces. > It is also independent of the standard implemented Un*x filesystems, > for which you will have to pay extra if you want to take extra > measures against power loss. Rather, it is worse. With a Unix filesystem you get quite defined semantics about what is written when. > The problem with e.g. e2fs, is that it is not robust enough if a CPU > fails. ext2fs doesn't even claim to have Unix filesystem semantics. Regards, Giles From pgsql-hackers-owner+M1795@postgresql.org Thu Dec 7 18:47:52 2000 Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by candle.pha.pa.us (8.9.0/8.9.0) with ESMTP id SAA09172 for ; Thu, 7 Dec 2000 18:47:52 -0500 (EST) Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by mail.postgresql.org (8.11.1/8.11.1) with SMTP id eB7NjFP10612; Thu, 7 Dec 2000 18:45:15 -0500 (EST) (envelope-from pgsql-hackers-owner+M1795@postgresql.org) Received: from thor.tht.net (thor.tht.net [209.47.145.4]) by mail.postgresql.org (8.11.1/8.11.1) with ESMTP id eB7N6BP08233 for ; Thu, 7 Dec 2000 18:06:11 -0500 (EST) (envelope-from bright@fw.wintelcom.net) Received: from fw.wintelcom.net (bright@ns1.wintelcom.net [209.1.153.20]) by thor.tht.net (8.9.3/8.9.3) with ESMTP id SAA97456 for ; Thu, 7 Dec 2000 18:57:32 GMT (envelope-from bright@fw.wintelcom.net) Received: (from bright@localhost) by fw.wintelcom.net (8.10.0/8.10.0) id eB7MvWE21269 for pgsql-hackers@postgresql.org; Thu, 7 Dec 2000 14:57:32 -0800 (PST) Date: Thu, 7 Dec 2000 14:57:32 -0800 From: Alfred Perlstein To: pgsql-hackers@postgresql.org Subject: [HACKERS] Patches with vacuum fixes available for 7.0.x Message-ID: <20001207145732.X16205@fw.wintelcom.net> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.2.5i Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org Status: ORr We recently had a very satisfactory contract completed by Vadim. Basically Vadim has been able to reduce the amount of time taken by a vacuum from 10-15 minutes down to under 10 seconds. We've been running with these patches under heavy load for about a week now without any problems except one: don't 'lazy' (new option for vacuum) a table which has just had an index created on it, or at least don't expect it to take any less time than a normal vacuum would. There's three patchsets and they are available at: http://people.freebsd.org/~alfred/vacfix/ complete diff: http://people.freebsd.org/~alfred/vacfix/v.diff only lazy vacuum option to speed up index vacuums: http://people.freebsd.org/~alfred/vacfix/vlazy.tgz only lazy vacuum option to only scan from start of modified data: http://people.freebsd.org/~alfred/vacfix/mnmb.tgz Although the patches are for 7.0.x I'm hoping that they can be forward ported (if Vadim hasn't done it already) to 7.1. enjoy! -- -Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org] "I have the heart of a child; I keep it in a jar on my desk." From pgsql-hackers-owner+M1809@postgresql.org Thu Dec 7 20:27:39 2000 Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by candle.pha.pa.us (8.9.0/8.9.0) with ESMTP id UAA11827 for ; Thu, 7 Dec 2000 20:27:38 -0500 (EST) Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by mail.postgresql.org (8.11.1/8.11.1) with SMTP id eB81PsP22362; Thu, 7 Dec 2000 20:25:54 -0500 (EST) (envelope-from pgsql-hackers-owner+M1809@postgresql.org) Received: from fw.wintelcom.net (ns1.wintelcom.net [209.1.153.20]) by mail.postgresql.org (8.11.1/8.11.1) with ESMTP id eB81JkP21783 for ; Thu, 7 Dec 2000 20:19:46 -0500 (EST) (envelope-from bright@fw.wintelcom.net) Received: (from bright@localhost) by fw.wintelcom.net (8.10.0/8.10.0) id eB81JwU25447; Thu, 7 Dec 2000 17:19:58 -0800 (PST) Date: Thu, 7 Dec 2000 17:19:58 -0800 From: Alfred Perlstein To: Tom Lane cc: pgsql-hackers@postgresql.org Subject: Re: [HACKERS] Patches with vacuum fixes available for 7.0.x Message-ID: <20001207171958.B16205@fw.wintelcom.net> References: <20001207145732.X16205@fw.wintelcom.net> <28791.976236143@sss.pgh.pa.us> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.2.5i In-Reply-To: <28791.976236143@sss.pgh.pa.us>; from tgl@sss.pgh.pa.us on Thu, Dec 07, 2000 at 07:42:23PM -0500 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org Status: OR * Tom Lane [001207 17:10] wrote: > Alfred Perlstein writes: > > Basically Vadim has been able to reduce the amount of time > > taken by a vacuum from 10-15 minutes down to under 10 seconds. > > Cool. What's it do, exactly? ================================================================ The first is a bonus that Vadim gave us to speed up index vacuums, I'm not sure I understand it completely, but it work really well. :) here's the README he gave us: Vacuum LAZY index cleanup option LAZY vacuum option introduces new way of indices cleanup. Instead of reading entire index file to remove index tuples pointing to deleted table records, with LAZY option vacuum performes index scans using keys fetched from table record to be deleted. Vacuum checks each result returned by index scan if it points to target heap record and removes corresponding index tuple. This can greatly speed up indices cleaning if not so many table records were deleted/modified between vacuum runs. Vacuum uses new option on user' demand. New vacuum syntax is: vacuum [verbose] [analyze] [lazy] [table [(columns)]] ================================================================ The second is one of the suggestions I gave on the lists a while back, keeping track of the "last dirtied" block in the data files to only scan the tail end of the file for deleted rows, I think what he instead did was keep a table that holds all the modified blocks and vacuum only scans those: Minimal Number Modified Block (MNMB) This feature is to track MNMB of required tables with triggers to avoid reading unmodified table pages by vacuum. Triggers store MNMB in per-table files in specified directory ($LIBDIR/contrib/mnmb by default) and create these files if not existed. Vacuum first looks up functions mnmb_getblock(Oid databaseId, Oid tableId) mnmb_setblock(Oid databaseId, Oid tableId, Oid block) in catalog. If *both* functions were found *and* there was no ANALYZE option specified then vacuum calls mnmb_getblock to obtain MNMB for table being vacuumed and starts reading this table from block number returned. After table was processed vacuum calls mnmb_setblock to update data in file to last table block number. Neither mnmb_getblock nor mnmb_setblock try to create file. If there was no file for table being vacuumed then mnmb_getblock returns 0 and mnmb_setblock does nothing. mnmb_setblock() may be used to set in file MNMB to 0 and force vacuum to read entire table if required. To compile MNMB you have to add -DMNMB to CUSTOM_COPT in src/Makefile.custom. -- -Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org] "I have the heart of a child; I keep it in a jar on my desk." From pgsql-general-owner+M4010@postgresql.org Mon Feb 5 18:50:47 2001 Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by candle.pha.pa.us (8.9.0/8.9.0) with ESMTP id SAA02209 for ; Mon, 5 Feb 2001 18:50:46 -0500 (EST) Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by mail.postgresql.org (8.11.1/8.11.1) with SMTP id f15Nn8x86486; Mon, 5 Feb 2001 18:49:08 -0500 (EST) (envelope-from pgsql-general-owner+M4010@postgresql.org) Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by mail.postgresql.org (8.11.1/8.11.1) with SMTP id f15N7Ux81124 for ; Mon, 5 Feb 2001 18:07:30 -0500 (EST) (envelope-from pgsql-general-owner@postgresql.org) Received: from news.tht.net (news.hub.org [216.126.91.242]) by mail.postgresql.org (8.11.1/8.11.1) with ESMTP id f0V0Twq69854 for ; Tue, 30 Jan 2001 19:29:58 -0500 (EST) (envelope-from news@news.tht.net) Received: (from news@localhost) by news.tht.net (8.11.1/8.11.1) id f0V0RAO01011 for pgsql-general@postgresql.org; Tue, 30 Jan 2001 19:27:10 -0500 (EST) (envelope-from news) From: Mike Hoskins X-Newsgroups: comp.databases.postgresql.general Subject: Re: [GENERAL] MySQL file system Date: Tue, 30 Jan 2001 18:30:36 -0600 Organization: Hub.Org Networking Services (http://www.hub.org) Lines: 120 Message-ID: <3A775CAB.C416AA16@yahoo.com> References: <016e01c080b7$ea554080$330a0a0a@6014cwpza006> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Complaints-To: scrappy@hub.org X-Mailer: Mozilla 4.76 [en] (Windows NT 5.0; U) X-Accept-Language: en To: pgsql-general@postgresql.org Precedence: bulk Sender: pgsql-general-owner@postgresql.org Status: OR This idea is such a popular (even old) one that Oracle developed it for 8i -- IFS. Yep, AS/400 has had it forever, and BeOS is another example. Informix has had its DataBlades for years, as well. In fact, Reiser-FS is an FS implemented on a DB, albeit probably not a SQL DB. AIX's LVM and JFS is extent/DB-based, as well. Let's see now, why would all those guys do that? (Now, some of those that aren't SQL-based probably won't allow SQL queries on files, so just think about those that do, for a minute).... Rather than asking why, a far better question is why not? There is SO much functionality to be gained here that it's silly to ask why. At a higher level, treating BLOBs as files and as DB entries simultaneously has so many uses, that one has trouble answering the question properly without the puzzled stare back at the questioner. Again, look at the above list, particularly at AS/400 -- the entire OS's FS sits on top of DB/2! For example, think how easy dynamically generated web sites could access online catalog information, with all those JPEG's, GIFs, PNGs, HTML files, Text files, .PDF's, etc., both in the DB and in the FS. This would be so much easier to maintain, when you have webmasters, web designers, artists, programmers, sysadmins, dba's, etc., all trying to manage a big, dynamic, graphics-rich web site. Who cares if the FS is a bit slow, as long as it's not too slow? That's not the point, anyway. The point is easy access to data: asset management, version control, the ability to access the same data as a file and as a BLOB simultaneously, the ability to replicate easier, the ability to use more tools on the same info, etc. It's not for speed, per se; instead, it's for accessibility. Think about this issue. You have some already compiled text-based program that works on binary files, but not on databases -- it was simply never designed into the program. How are you going to get your graphics BLOBs into that program? Oh yeah, let's write another program to transform our data into files, first, then after processing delete them in some cleanup routine.... Why? If you have a DB'ed FS, then file data can simultaneously have two views -- one for the DB and one as an FS. (You can easily reverse the scenario.) Not only does this save time and disk space; it saves you from having to pay for the most expensive element of all -- programmer time. BTW, once this FS-on-a-DB concept really sinks in, imagine how tightly integrated Linux/Unix apps could be written. Imagine if a bunch of GPL'ed software started coding for this and used this as a means to exchange data, all using a common set of libraries. You could get to the point of uniting files, BLOBs, data of all sorts, IPC, version control, etc., all under one umbrella, especially if XML was the means data was exchanged. Heck, distributed authentication, file access, data access, etc., could be improved greatly. Well, this paragraph sounds like flame bait, but really consider the ramifications. Also, read the next paragraph.... Something like this *has* existed for Postgres for a long time -- PGFS, by Brian Bartholomew. It's even supposedly matured with age. Unfortunately, I cannot get to http://www.wv.com/ (Working Version's main site). Working Version is a version control system that keeps old versions of files around in the FS. It uses PG as the back-end DB and lets you mount it like another FS. It's supposedly an awesome system, but where is it? It's not some clunky korbit thingy, either. (If someone can find it, please let me know by email, if possible.) The only thing I can find on this is from a Google search, which caches everything but the actual software: http://www.google.com/search?q=pgfs+postgres&num=100&hl=en&lr=lang_en&newwindow=1&safe=active Also, there is the Perl-FS that can be transformed into something like PGFS: http://www.assurdo.com/perlfs/ It allows you to write Perl code that can mount various protocols or data types as an FS, in user space. (One example is the ability to mount FTP sites, BTW.) Instead of ridiculing something you've never tried, consider that MySQL-FS, Oracle (IFS), Informix (DataBlades), AS/400 (DB/2), BeOS, and Reiser-FS are doing this today. Do you want to be left behind and let them tell us what it's good for? Or, do we want this for PG? (Reiser-FS, BTW, is FASTER than ext2, but has no SQL hooks). There were many posts on this on slashdot: http://slashdot.org/article.pl?sid=01/01/16/1855253&mode=thread (I wrote some comments here, as well, just look for mikehoskins) I, for one, want to see this succeed for MySQL, PostgreSQL, msql, etc. It's an awesome feature that doesn't need to be speedy because it can save HUMANS time. The question really is, "When do we want to catch up to everyone else?" We are always moving to higher levels of abstraction, anyway, so it's just a matter of time. PG should participate. Adam Lang wrote: > I wasn't following the thread too closely, but database for a filesystem has > been done. BeOS uses a database for a filesystem as well as AS/400 and > Mainframes. > > Adam Lang > Systems Engineer > Rutgers Casualty Insurance Company > http://www.rutgersinsurance.com > ----- Original Message ----- > From: "Alfred Perlstein" > To: "Robert D. Nelson" > Cc: "Joseph Shraibman" ; "Karl DeBisschop" > ; "Ned Lilly" ; "PostgreSQL > General" > Sent: Wednesday, January 17, 2001 12:23 PM > Subject: Re: [GENERAL] MySQL file system > > > * Robert D. Nelson [010117 05:17] wrote: > > > >Raw disk access allows: > > > > > > If I'm correct, mysql is providing a filesystem, not a way to access raw > > > disk, like Oracle does. Huge difference there - with a filesystem, you > have > > > overhead of FS *and* SQL at the same time. > > > > Oh, so it's sort of like /proc for mysql? > > > > What a terrible waste of time and resources. :( > > > > -- > > -Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org] > > "I have the heart of a child; I keep it in a jar on my desk." From pgsql-general-owner+M4049@postgresql.org Tue Feb 6 01:26:19 2001 Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by candle.pha.pa.us (8.9.0/8.9.0) with ESMTP id BAA21425 for ; Tue, 6 Feb 2001 01:26:18 -0500 (EST) Received: from mail.postgresql.org (webmail.postgresql.org [216.126.85.28]) by mail.postgresql.org (8.11.1/8.11.1) with SMTP id f166Nxx26400; Tue, 6 Feb 2001 01:23:59 -0500 (EST) (envelope-from pgsql-general-owner+M4049@postgresql.org) Received: from simecity.com ([202.188.254.2]) by mail.postgresql.org (8.11.1/8.11.1) with ESMTP id f166GUx25754 for ; Tue, 6 Feb 2001 01:16:30 -0500 (EST) (envelope-from lyeoh@pop.jaring.my) Received: (from mail@localhost) by simecity.com (8.9.3/8.8.7) id OAA23910; Tue, 6 Feb 2001 14:28:48 +0800 Received: from (ilab2.mecomb.po.my [192.168.3.22]) by cirrus.simecity.com via smap (V2.1) id xma023908; Tue, 6 Feb 01 14:28:34 +0800 Message-ID: <3.0.5.32.20010206141555.00a3d100@192.228.128.13> X-Sender: lyeoh@192.228.128.13 X-Mailer: QUALCOMM Windows Eudora Light Version 3.0.5 (32) Date: Tue, 06 Feb 2001 14:15:55 +0800 To: Mike Hoskins , pgsql-general@postgresql.org From: Lincoln Yeoh Subject: [GENERAL] Re: MySQL file system In-Reply-To: <3A775CF7.3C5F1909@yahoo.com> References: <016e01c080b7$ea554080$330a0a0a@6014cwpza006> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Precedence: bulk Sender: pgsql-general-owner@postgresql.org Status: OR What you're saying seems to be to have a data structure where the same data can be accessed in both the filesystem style and the RDBMs style. How does that work? How is the mapping done between both structures? Slapping a filesystem on top of a RDBMs doesn't do that does it? Most filesystems are basically databases already, just differently structured and featured databases. And so far most of them do their job pretty well. You move a folder/directory somewhere, and everything inside it moves. Tons of data are already arranged in that form. Though porting over data from one filesystem to another is not always straightforward, RDBMSes are far worse. Maybe what would be nice is not a filesystem based on a database, rather one influenced by databases. One with a decent fulltextindex for data and filenames, where you have the option to ignore or not ignore nonalphanumerics and still get an indexed search. Then perhaps we could do something like the following: select file.name from path "/var/logs/" where file.name like "%.log%' and file.lastmodified > '2000/1/1' and file.contents =~ 'te_st[0-9]+\.gif$' use index Checkpoints would be nice too. Then I can rollback to a known point if I screw up ;). In fact the SQL style interface doesn't have to be built in at all. Neither does the index have to be realtime. I suppose there could be an option to make it realtime if performance is not an issue. What could be done is to use some fast filesystem. Then we add tools to maintain indexes, for SQL style interfaces and other style interfaces. Checkpoints and rollbacks would be harder of course. Cheerio, Link. From pgsql-hackers-owner+M20329@postgresql.org Tue Mar 19 18:00:15 2002 Return-path: Received: from postgresql.org (postgresql.org [64.49.215.8]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g2K00EA02465 for ; Tue, 19 Mar 2002 19:00:14 -0500 (EST) Received: from postgresql.org (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with SMTP id 8C7164763EF; Tue, 19 Mar 2002 18:22:08 -0500 (EST) Received: from CopelandConsulting.Net (dsl-24293-ld.customer.centurytel.net [209.142.135.135]) by postgresql.org (Postfix) with ESMTP id E4DAD475F1F for ; Tue, 19 Mar 2002 18:02:17 -0500 (EST) Received: from mouse.copelandconsulting.net (mouse.copelandconsulting.net [192.168.1.2]) by CopelandConsulting.Net (8.10.1/8.10.1) with ESMTP id g2JN0jh13185; Tue, 19 Mar 2002 17:00:45 -0600 (CST) X-Trade-Id: To: Matthew Kirkwood cc: Oleg Bartunov , PostgresSQL Hackers Mailing List Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="=-Ivchb84S75fOMzJ9DxwK" X-Mailer: Evolution/1.0.2 Date: 19 Mar 2002 17:00:53 -0600 Message-ID: <1016578854.14670.450.camel@mouse.copelandconsulting.net> MIME-Version: 1.0 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org Status: OR --=-Ivchb84S75fOMzJ9DxwK Content-Type: text/plain Content-Transfer-Encoding: quoted-printable On Tue, 2002-03-19 at 15:30, Matthew Kirkwood wrote: > On Tue, 19 Mar 2002, Oleg Bartunov wrote: >=20 > Sorry to reply over you, Oleg. >=20 > > On 13 Mar 2002, Greg Copeland wrote: > > > > > One of the reasons why I originally stated following the hackers list= is > > > because I wanted to implement bitmap indexes. I found in the archive= s, > > > the follow link, http://www.it.iitb.ernet.in/~rvijay/dbms/proj/, which > > > was extracted from this, > > > http://groups.google.com/groups?hl=3Den&threadm=3D01C0EF67.5105D2E0.m= ascarm%40mascari.com&rnum=3D1&prev=3D/groups%3Fq%3Dbitmap%2Bindex%2Bgroup:c= omp.databases.postgresql.hackers%26hl%3Den%26selm%3D01C0EF67.5105D2E0.masca= rm%2540mascari.com%26rnum%3D1, archive thread. >=20 > For every case I have used a bitmap index on Oracle, a > partial index[0] made more sense (especialy since it > could usefully be compound). That's very true, however, often bitmap indexes are used where partial indexes may not work well. It maybe you were trying to apply the cure for the wrong disease. ;) >=20 > Our troublesome case (on Oracle) is a table of "events" > where maybe fifty to a couple of hundred are "published" > (ie. web-visible) at any time. The events are categorised > by sport (about a dozen) and by "event type" (about five). > We never really query events except by PK or by sport/type/ > published. The reason why bitmap indexes are primarily used for DSS and data wherehousing applications is because they are best used on extremely large to very large tables which have low cardinality (e.g, 10,000,000 rows having 200 distinct values). On top of that, bitmap indexes also tend to be much smaller than their "standard" cousins. On large and very tables tables, this can sometimes save gigs in index space alone (serious space benefit). Plus, their small index size tends to result in much less I/O (serious speed benefit). This, of course, can result in several orders of magnitude speed improvements when index scans are required. As an added bonus, using AND, OR, XOR and NOT predicates are exceptionally fast and if implemented properly, can even take advantage of some 64-bit hardware for further speed improvements. This, of course, further speeds look ups. The primary down side is that inserts and updates to bitmap indexes are very costly (comparatively) which is, yet again, why they excel in read-only environments (DSS & data wherehousing). It should also be noted that RDMS's, such as Oracle, often use multiple types of bitmap indexes. This further impedes insert/update performance, however, the additional bitmap index types usually allow for range predicates while still making use of the bitmap index. If I'm not mistaken, several other types of bitmaps are available as well as many ways to encode and compress (rle, quad compression, etc) bitmap indexes which further save on an already compact indexing scheme. Given the proper problem domain, index bitmaps can be a big win. >=20 > We make a bitmap index on "published", and trust Oracle to > use it correctly, and hope that our other indexes are also > useful. >=20 > On Postgres[1] we would make a partial compound index: >=20 > create index ... on events(sport_id,event_type_id) > where published=3D'Y'; Generally speaking, bitmap indexes will not serve you very will on tables having a low row counts, high cardinality or where they are attached to tables which are primarily used in an OLTP capacity.=20 Situations where you have a low row count and low cardinality or high row count and high cardinality tend to be better addressed by partial indexes; which seem to make much more sense. In your example, it sounds like you did "the right thing"(tm). ;) Greg --=-Ivchb84S75fOMzJ9DxwK Content-Type: application/pgp-signature; name=signature.asc Content-Description: This is a digitally signed message part -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.6 (GNU/Linux) Comment: For info see http://www.gnupg.org iD8DBQA8l8Ml4lr1bpbcL6kRAhldAJ9Aoi9dwm1OteZjySfsd1o42trWLACfegQj OEV6eO8MnBSlbJMHiQ08gNE= =PQvW -----END PGP SIGNATURE----- --=-Ivchb84S75fOMzJ9DxwK-- From pgsql-hackers-owner+M26157@postgresql.org Tue Aug 6 23:06:34 2002 Date: Wed, 7 Aug 2002 13:07:38 +1000 (EST) From: Gavin Sherry To: Curt Sampson cc: pgsql-hackers@postgresql.org Subject: Re: [HACKERS] CLUSTER and indisclustered In-Reply-To: Message-ID: X-Virus-Scanned: by AMaViS new-20020517 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org X-Virus-Scanned: by AMaViS new-20020517 Content-Length: 1357 On Wed, 7 Aug 2002, Curt Sampson wrote: > But after doing some benchmarking of various sorts of random reads > and writes, it occurred to me that there might be optimizations > that could help a lot with this sort of thing. What if, when we've > got an index block with a bunch of entries, instead of doing the > reads in the order of the entries, we do them in the order of the > blocks the entries point to? That would introduce a certain amount > of "sequentialness" to the reads that the OS is not capable of > introducing (since it can't reschedule the reads you're doing, the > way it could reschedule, say, random writes). This sounds more or less like the method employed by Firebird as described by Ann Douglas to Tom at OSCON (correct me if I get this wrong). Basically, firebird populates a bitmap with entries the scan is interested in. The bitmap is populated in page order so that all entries on the same heap page can be fetched at once. This is totally different to the way postgres does things and would require significant modification to the index access methods. Gavin ---------------------------(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 From pgsql-hackers-owner+M26162@postgresql.org Wed Aug 7 00:42:35 2002 To: Curt Sampson cc: mark Kirkwood , Gavin Sherry , Bruce Momjian , pgsql-hackers@postgresql.org Subject: Re: [HACKERS] CLUSTER and indisclustered In-Reply-To: References: Comments: In-reply-to Curt Sampson message dated "Wed, 07 Aug 2002 11:31:32 +0900" Date: Wed, 07 Aug 2002 00:41:47 -0400 Message-ID: <12593.1028695307@sss.pgh.pa.us> From: Tom Lane X-Virus-Scanned: by AMaViS new-20020517 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org X-Virus-Scanned: by AMaViS new-20020517 Content-Length: 3063 Curt Sampson writes: > But after doing some benchmarking of various sorts of random reads > and writes, it occurred to me that there might be optimizations > that could help a lot with this sort of thing. What if, when we've > got an index block with a bunch of entries, instead of doing the > reads in the order of the entries, we do them in the order of the > blocks the entries point to? I thought to myself "didn't I just post something about that?" and then realized it was on a different mailing list. Here ya go (and no, this is not the first time around on this list either...) I am currently thinking that bitmap indexes per se are not all that interesting. What does interest me is bitmapped index lookup, which came back into mind after hearing Ann Harrison describe how FireBird/ InterBase does it. The idea is that you don't scan the index and base table concurrently as we presently do it. Instead, you scan the index and make a list of the TIDs of the table tuples you need to visit. This list can be conveniently represented as a sparse bitmap. After you've finished looking at the index, you visit all the required table tuples *in physical order* using the bitmap. This eliminates multiple fetches of the same heap page, and can possibly let you get some win from sequential access. Once you have built this mechanism, you can then move on to using multiple indexes in interesting ways: you can do several indexscans in one query and then AND or OR their bitmaps before doing the heap scan. This would allow, for example, "WHERE a = foo and b = bar" to be handled by ANDing results from separate indexes on the a and b columns, rather than having to choose only one index to use as we do now. Some thoughts about implementation: FireBird's implementation seems to depend on an assumption about a fixed number of tuple pointers per page. We don't have that, but we could probably get away with just allocating BLCKSZ/sizeof(HeapTupleHeaderData) bits per page. Also, the main downside of this approach is that the bitmap could get large --- but you could have some logic that causes you to fall back to plain sequential scan if you get too many index hits. (It's interesting to think of this as lossy compression of the bitmap... which leads to the idea of only being fuzzy in limited areas of the bitmap, rather than losing all the information you have.) A possibly nasty issue is that lazy VACUUM has some assumptions in it about indexscans holding pins on index pages --- that's what prevents it from removing heap tuples that a concurrent indexscan is just about to visit. It might be that there is no problem: even if lazy VACUUM removes a heap tuple and someone else then installs a new tuple in that same TID slot, you should be okay because the new tuple is too new to pass your visibility test. But I'm not convinced this is safe. regards, tom lane ---------------------------(end of broadcast)--------------------------- TIP 6: Have you searched our list archives? http://archives.postgresql.org From pgsql-hackers-owner+M26172@postgresql.org Wed Aug 7 02:49:56 2002 X-Authentication-Warning: rh72.home.ee: hannu set sender to hannu@tm.ee using -f Subject: Re: [HACKERS] CLUSTER and indisclustered From: Hannu Krosing To: Tom Lane cc: Curt Sampson , mark Kirkwood , Gavin Sherry , Bruce Momjian , pgsql-hackers@postgresql.org In-Reply-To: <12776.1028697148@sss.pgh.pa.us> References: <12776.1028697148@sss.pgh.pa.us> X-Mailer: Ximian Evolution 1.0.7 Date: 07 Aug 2002 09:46:29 +0500 Message-ID: <1028695589.2133.11.camel@rh72.home.ee> X-Virus-Scanned: by AMaViS new-20020517 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org X-Virus-Scanned: by AMaViS new-20020517 Content-Length: 1064 On Wed, 2002-08-07 at 10:12, Tom Lane wrote: > Curt Sampson writes: > > On Wed, 7 Aug 2002, Tom Lane wrote: > >> Also, the main downside of this approach is that the bitmap could > >> get large --- but you could have some logic that causes you to fall > >> back to plain sequential scan if you get too many index hits. > > > Well, what I was thinking of, should the list of TIDs to fetch get too > > long, was just to break it down in to chunks. > > But then you lose the possibility of combining multiple indexes through > bitmap AND/OR steps, which seems quite interesting to me. If you've > visited only a part of each index then you can't apply that concept. When the tuples are small relative to pagesize, you may get some "compression" by saving just pages and not the actual tids in the the bitmap. ------------- Hannu ---------------------------(end of broadcast)--------------------------- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to majordomo@postgresql.org) From pgsql-hackers-owner+M26166@postgresql.org Wed Aug 7 00:55:52 2002 Date: Wed, 7 Aug 2002 13:55:41 +0900 (JST) From: Curt Sampson To: Tom Lane cc: mark Kirkwood , Gavin Sherry , Bruce Momjian , Subject: Re: [HACKERS] CLUSTER and indisclustered In-Reply-To: <12593.1028695307@sss.pgh.pa.us> Message-ID: X-Virus-Scanned: by AMaViS new-20020517 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org X-Virus-Scanned: by AMaViS new-20020517 Content-Length: 1840 On Wed, 7 Aug 2002, Tom Lane wrote: > I thought to myself "didn't I just post something about that?" > and then realized it was on a different mailing list. Here ya go > (and no, this is not the first time around on this list either...) Wow. I'm glad to see you looking at this, because this feature would so *so* much for the performance of some of my queries, and really, really impress my "billion-row-database" client. > The idea is that you don't scan the index and base table concurrently > as we presently do it. Instead, you scan the index and make a list > of the TIDs of the table tuples you need to visit. Right. > Also, the main downside of this approach is that the bitmap could > get large --- but you could have some logic that causes you to fall > back to plain sequential scan if you get too many index hits. Well, what I was thinking of, should the list of TIDs to fetch get too long, was just to break it down in to chunks. If you want to limit to, say, 1000 TIDs, and your index has 3000, just do the first 1000, then the next 1000, then the last 1000. This would still result in much less disk head movement and speed the query immensely. (BTW, I have verified this emperically during testing of random read vs. random write on a RAID controller. The writes were 5-10 times faster than the reads because the controller was caching a number of writes and then doing them in the best possible order, whereas the reads had to be satisfied in the order they were submitted to the controller.) cjs -- Curt Sampson +81 90 7737 2974 http://www.netbsd.org Don't you know, in this new Dark Age, we're all light. --XTC ---------------------------(end of broadcast)--------------------------- TIP 5: Have you checked our extensive FAQ? http://www.postgresql.org/users-lounge/docs/faq.html From pgsql-hackers-owner+M26167@postgresql.org Wed Aug 7 01:12:54 2002 To: Curt Sampson cc: mark Kirkwood , Gavin Sherry , Bruce Momjian , pgsql-hackers@postgresql.org Subject: Re: [HACKERS] CLUSTER and indisclustered In-Reply-To: References: Comments: In-reply-to Curt Sampson message dated "Wed, 07 Aug 2002 13:55:41 +0900" Date: Wed, 07 Aug 2002 01:12:28 -0400 Message-ID: <12776.1028697148@sss.pgh.pa.us> From: Tom Lane X-Virus-Scanned: by AMaViS new-20020517 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org X-Virus-Scanned: by AMaViS new-20020517 Content-Length: 1428 Curt Sampson writes: > On Wed, 7 Aug 2002, Tom Lane wrote: >> Also, the main downside of this approach is that the bitmap could >> get large --- but you could have some logic that causes you to fall >> back to plain sequential scan if you get too many index hits. > Well, what I was thinking of, should the list of TIDs to fetch get too > long, was just to break it down in to chunks. But then you lose the possibility of combining multiple indexes through bitmap AND/OR steps, which seems quite interesting to me. If you've visited only a part of each index then you can't apply that concept. Another point to keep in mind is that the bigger the bitmap gets, the less useful an indexscan is, by definition --- sooner or later you might as well fall back to a seqscan. So the idea of lossy compression of a large bitmap seems really ideal to me. In principle you could seqscan the parts of the table where matching tuples are thick on the ground, and indexscan the parts where they ain't. Maybe this seems natural to me as an old JPEG campaigner, but if you don't see the logic I recommend thinking about it a little ... regards, tom lane ---------------------------(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 From tgl@sss.pgh.pa.us Wed Aug 7 09:27:05 2002 To: Hannu Krosing cc: Curt Sampson , mark Kirkwood , Gavin Sherry , Bruce Momjian , pgsql-hackers@postgresql.org Subject: Re: [HACKERS] CLUSTER and indisclustered In-Reply-To: <1028726966.13418.12.camel@taru.tm.ee> References: <12776.1028697148@sss.pgh.pa.us> <1028695589.2133.11.camel@rh72.home.ee> <1028726966.13418.12.camel@taru.tm.ee> Comments: In-reply-to Hannu Krosing message dated "07 Aug 2002 15:29:26 +0200" Date: Wed, 07 Aug 2002 09:26:42 -0400 Message-ID: <15010.1028726802@sss.pgh.pa.us> From: Tom Lane Content-Length: 1120 Hannu Krosing writes: > Now I remembered my original preference for page bitmaps (vs. tuple > bitmaps): one can't actually make good use of a bitmap of tuples because > there is no fixed tuples/page ratio and thus no way to quickly go from > bit position to actual tuple. You mention the same problem but propose a > different solution. > Using page bitmap, we will at least avoid fetching any unneeded pages - > essentially we will have a sequential scan over possibly interesting > pages. Right. One form of the "lossy compression" idea I suggested is to switch from a per-tuple bitmap to a per-page bitmap once the bitmap gets too large to work with. Again, one could imagine doing that only in denser areas of the bitmap. > But I guess that CLUSTER support for INSERT will not be touched for 7.3 > as will real bitmap indexes ;) All of this is far-future work I think. Adding a new scan type to the executor would probably be pretty localized, but the ramifications in the planner could be extensive --- especially if you want to do plans involving ANDed or ORed bitmaps. regards, tom lane From pgsql-hackers-owner+M26178@postgresql.org Wed Aug 7 08:28:14 2002 X-Authentication-Warning: taru.tm.ee: hannu set sender to hannu@tm.ee using -f Subject: Re: [HACKERS] CLUSTER and indisclustered From: Hannu Krosing To: Hannu Krosing cc: Tom Lane , Curt Sampson , mark Kirkwood , Gavin Sherry , Bruce Momjian , pgsql-hackers@postgresql.org In-Reply-To: <1028695589.2133.11.camel@rh72.home.ee> References: <12776.1028697148@sss.pgh.pa.us> <1028695589.2133.11.camel@rh72.home.ee> X-Mailer: Ximian Evolution 1.0.3.99 Date: 07 Aug 2002 15:29:26 +0200 Message-ID: <1028726966.13418.12.camel@taru.tm.ee> X-Virus-Scanned: by AMaViS new-20020517 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org X-Virus-Scanned: by AMaViS new-20020517 Content-Length: 1837 On Wed, 2002-08-07 at 06:46, Hannu Krosing wrote: > On Wed, 2002-08-07 at 10:12, Tom Lane wrote: > > Curt Sampson writes: > > > On Wed, 7 Aug 2002, Tom Lane wrote: > > >> Also, the main downside of this approach is that the bitmap could > > >> get large --- but you could have some logic that causes you to fall > > >> back to plain sequential scan if you get too many index hits. > > > > > Well, what I was thinking of, should the list of TIDs to fetch get too > > > long, was just to break it down in to chunks. > > > > But then you lose the possibility of combining multiple indexes through > > bitmap AND/OR steps, which seems quite interesting to me. If you've > > visited only a part of each index then you can't apply that concept. > > When the tuples are small relative to pagesize, you may get some > "compression" by saving just pages and not the actual tids in the the > bitmap. Now I remembered my original preference for page bitmaps (vs. tuple bitmaps): one can't actually make good use of a bitmap of tuples because there is no fixed tuples/page ratio and thus no way to quickly go from bit position to actual tuple. You mention the same problem but propose a different solution. Using page bitmap, we will at least avoid fetching any unneeded pages - essentially we will have a sequential scan over possibly interesting pages. If we were to use page-bitmap index for something with only a few values like booleans, some insert-time local clustering should be useful, so that TRUEs and FALSEs end up on different pages. But I guess that CLUSTER support for INSERT will not be touched for 7.3 as will real bitmap indexes ;) --------------- Hannu ---------------------------(end of broadcast)--------------------------- TIP 6: Have you searched our list archives? http://archives.postgresql.org From pgsql-hackers-owner+M26192@postgresql.org Wed Aug 7 10:26:30 2002 To: Hannu Krosing cc: Curt Sampson , mark Kirkwood , Gavin Sherry , Bruce Momjian , pgsql-hackers@postgresql.org Subject: Re: [HACKERS] CLUSTER and indisclustered In-Reply-To: <1028733234.13418.113.camel@taru.tm.ee> References: <12776.1028697148@sss.pgh.pa.us> <1028695589.2133.11.camel@rh72.home.ee> <1028726966.13418.12.camel@taru.tm.ee> <15010.1028726802@sss.pgh.pa.us> <1028733234.13418.113.camel@taru.tm.ee> Comments: In-reply-to Hannu Krosing message dated "07 Aug 2002 17:13:54 +0200" Date: Wed, 07 Aug 2002 10:26:13 -0400 Message-ID: <15622.1028730373@sss.pgh.pa.us> From: Tom Lane X-Virus-Scanned: by AMaViS new-20020517 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org X-Virus-Scanned: by AMaViS new-20020517 Content-Length: 1224 Hannu Krosing writes: > On Wed, 2002-08-07 at 15:26, Tom Lane wrote: >> Right. One form of the "lossy compression" idea I suggested is to >> switch from a per-tuple bitmap to a per-page bitmap once the bitmap gets >> too large to work with. > If it is a real bitmap, should it not be easyeast to allocate at the > start ? But it isn't a "real bitmap". That would be a really poor implementation, both for space and speed --- do you really want to scan over a couple of megs of zeroes to find the few one-bits you care about, in the typical case? "Bitmap" is a convenient term because it describes the abstract behavior we want, but the actual data structure will probably be nontrivial. If I recall Ann's description correctly, Firebird's implementation uses run length coding of some kind (anyone care to dig in their source and get all the details?). If we tried anything in the way of lossy compression then there'd be even more stuff lurking under the hood. regards, tom lane ---------------------------(end of broadcast)--------------------------- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to majordomo@postgresql.org) From pgsql-hackers-owner+M26188@postgresql.org Wed Aug 7 10:12:26 2002 X-Authentication-Warning: taru.tm.ee: hannu set sender to hannu@tm.ee using -f Subject: Re: [HACKERS] CLUSTER and indisclustered From: Hannu Krosing To: Tom Lane cc: Curt Sampson , mark Kirkwood , Gavin Sherry , Bruce Momjian , pgsql-hackers@postgresql.org In-Reply-To: <15010.1028726802@sss.pgh.pa.us> References: <12776.1028697148@sss.pgh.pa.us> <1028695589.2133.11.camel@rh72.home.ee> <1028726966.13418.12.camel@taru.tm.ee> <15010.1028726802@sss.pgh.pa.us> X-Mailer: Ximian Evolution 1.0.3.99 Date: 07 Aug 2002 17:13:54 +0200 Message-ID: <1028733234.13418.113.camel@taru.tm.ee> X-Virus-Scanned: by AMaViS new-20020517 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org X-Virus-Scanned: by AMaViS new-20020517 Content-Length: 2812 On Wed, 2002-08-07 at 15:26, Tom Lane wrote: > Hannu Krosing writes: > > Now I remembered my original preference for page bitmaps (vs. tuple > > bitmaps): one can't actually make good use of a bitmap of tuples because > > there is no fixed tuples/page ratio and thus no way to quickly go from > > bit position to actual tuple. You mention the same problem but propose a > > different solution. > > > Using page bitmap, we will at least avoid fetching any unneeded pages - > > essentially we will have a sequential scan over possibly interesting > > pages. > > Right. One form of the "lossy compression" idea I suggested is to > switch from a per-tuple bitmap to a per-page bitmap once the bitmap gets > too large to work with. If it is a real bitmap, should it not be easyeast to allocate at the start ? a page bitmap for a 100 000 000 tuple table with 10 tuples/page will be sized 10000000/8 = 1.25 MB, which does not look too big for me for that amount of data (the data table itself would occupy 80 GB). Even having the bitmap of 16 bits/page (with the bits 0-14 meaning tuples 0-14 and bit 15 meaning "seq scan the rest of page") would consume just 20 MB of _local_ memory, and would be quite justifyiable for a query on a table that large. For a real bitmap index the tuples-per-page should be a user-supplied tuning parameter. > Again, one could imagine doing that only in denser areas of the bitmap. I would hardly call the resulting structure "a bitmap" ;) And I'm not sure the overhead for a more complex structure would win us any additional performance for most cases. > > But I guess that CLUSTER support for INSERT will not be touched for 7.3 > > as will real bitmap indexes ;) > > All of this is far-future work I think. After we do that we will probably be able claim support for "datawarehousing" ;) > Adding a new scan type to the > executor would probably be pretty localized, but the ramifications in > the planner could be extensive --- especially if you want to do plans > involving ANDed or ORed bitmaps. Also going to "smart inserter" which can do local clustering on sets of real bitmap indexes for INSERTS (and INSERT side of UPDATE) would probably be a major change from our current "stupid inserter" ;) This will not be needed for bitmap resolution higher than 1bit/page but default local clustering on bitmap indexes will probably buy us some extra performance. by avoiding data page fetches when such indexes are used. AN anyway the support for INSERT being aware of clustering will probably come up sometime. ------------ Hannu ---------------------------(end of broadcast)--------------------------- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to majordomo@postgresql.org) From hannu@tm.ee Wed Aug 7 11:22:53 2002 X-Authentication-Warning: taru.tm.ee: hannu set sender to hannu@tm.ee using -f Subject: Re: [HACKERS] CLUSTER and indisclustered From: Hannu Krosing To: Tom Lane cc: Curt Sampson , mark Kirkwood , Gavin Sherry , Bruce Momjian , pgsql-hackers@postgresql.org In-Reply-To: <15622.1028730373@sss.pgh.pa.us> References: <12776.1028697148@sss.pgh.pa.us> <1028695589.2133.11.camel@rh72.home.ee> <1028726966.13418.12.camel@taru.tm.ee> <15010.1028726802@sss.pgh.pa.us> <1028733234.13418.113.camel@taru.tm.ee> <15622.1028730373@sss.pgh.pa.us> X-Mailer: Ximian Evolution 1.0.3.99 Date: 07 Aug 2002 18:24:30 +0200 Message-ID: <1028737470.13419.182.camel@taru.tm.ee> Content-Length: 2382 On Wed, 2002-08-07 at 16:26, Tom Lane wrote: > Hannu Krosing writes: > > On Wed, 2002-08-07 at 15:26, Tom Lane wrote: > >> Right. One form of the "lossy compression" idea I suggested is to > >> switch from a per-tuple bitmap to a per-page bitmap once the bitmap gets > >> too large to work with. > > > If it is a real bitmap, should it not be easyeast to allocate at the > > start ? > > But it isn't a "real bitmap". That would be a really poor > implementation, both for space and speed --- do you really want to scan > over a couple of megs of zeroes to find the few one-bits you care about, > in the typical case? I guess that depends on data. The typical case should be somthing the stats process will find out so the optimiser can use it The bitmap must be less than 1/48 (size of TID) full for best uncompressed "active-tid-list" to be smaller than plain bitmap. If there were some structure above list then this ratio would be even higher. I have had good experience using "compressed delta lists", which will scale well ofer the whole "fullness" spectrum of bitmap, but this is for storage, not for initial constructing of lists. > "Bitmap" is a convenient term because it describes > the abstract behavior we want, but the actual data structure will > probably be nontrivial. If I recall Ann's description correctly, > Firebird's implementation uses run length coding of some kind (anyone > care to dig in their source and get all the details?). Plain RLL is probably a good way to store it and for merging two or more bitmaps, but not as good for constructing it bit-by-bit. I guess the most effective structure for updating is often still a plain bitmap (maybe not if it is very sparse and all of it does not fit in cache), followed by some kind of balanced tree (maybe rb-tree). If the bitmap is relatively full then the plain bitmap is almost always the most effective to update. > If we tried anything in the way of lossy compression then there'd > be even more stuff lurking under the hood. Having three-valued (0,1,maybe) RLL-encoded "tritmap" would be a good way to represent lossy compression, and it would also be quite straightforward to merge two of these using AND or OR. It may even be possible to easily construct it using a fixed-length b-tree and going from 1 to "maybe" for nodes that get too dense. --------------- Hannu From pgsql-hackers-owner+M21991@postgresql.org Wed Apr 24 23:37:37 2002 Return-path: Received: from postgresql.org (postgresql.org [64.49.215.8]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3P3ba416337 for ; Wed, 24 Apr 2002 23:37:36 -0400 (EDT) Received: from postgresql.org (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with SMTP id CF13447622B; Wed, 24 Apr 2002 23:37:31 -0400 (EDT) Received: from sraigw.sra.co.jp (sraigw.sra.co.jp [202.32.10.2]) by postgresql.org (Postfix) with ESMTP id 3EE92474E4B for ; Wed, 24 Apr 2002 23:37:19 -0400 (EDT) Received: from srascb.sra.co.jp (srascb [133.137.8.65]) by sraigw.sra.co.jp (8.9.3/3.7W-sraigw) with ESMTP id MAA76393; Thu, 25 Apr 2002 12:35:44 +0900 (JST) Received: (from root@localhost) by srascb.sra.co.jp (8.11.6/8.11.6) id g3P3ZCK64299; Thu, 25 Apr 2002 12:35:12 +0900 (JST) (envelope-from t-ishii@sra.co.jp) Received: from sranhm.sra.co.jp (sranhm [133.137.170.62]) by srascb.sra.co.jp (8.11.6/8.11.6av) with ESMTP id g3P3ZBV64291; Thu, 25 Apr 2002 12:35:11 +0900 (JST) (envelope-from t-ishii@sra.co.jp) Received: from localhost (IDENT:t-ishii@srapc1474.sra.co.jp [133.137.170.59]) by sranhm.sra.co.jp (8.9.3+3.2W/3.7W-srambox) with ESMTP id MAA25562; Thu, 25 Apr 2002 12:35:43 +0900 To: tgl@sss.pgh.pa.us cc: cjs@cynic.net, pgman@candle.pha.pa.us, pgsql-hackers@postgresql.org Subject: Re: [HACKERS] Sequential Scan Read-Ahead In-Reply-To: <12342.1019705420@sss.pgh.pa.us> References: <12342.1019705420@sss.pgh.pa.us> X-Mailer: Mew version 1.94.2 on Emacs 20.7 / Mule 4.1 =?iso-2022-jp?B?KBskQjAqGyhCKQ==?= MIME-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <20020425123429E.t-ishii@sra.co.jp> Date: Thu, 25 Apr 2002 12:34:29 +0900 From: Tatsuo Ishii X-Dispatcher: imput version 20000228(IM140) Lines: 12 Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org Status: OR > Curt Sampson writes: > > Grabbing bigger chunks is always optimal, AFICT, if they're not > > *too* big and you use the data. A single 64K read takes very little > > longer than a single 8K read. > > Proof? Long time ago I tested with the 32k block size and got 1.5-2x speed up comparing ordinary 8k block size in the sequential scan case. FYI, if this is the case. -- Tatsuo Ishii ---------------------------(end of broadcast)--------------------------- TIP 5: Have you checked our extensive FAQ? http://www.postgresql.org/users-lounge/docs/faq.html From mloftis@wgops.com Thu Apr 25 01:43:14 2002 Return-path: Received: from free.wgops.com (root@dsl092-002-178.sfo1.dsl.speakeasy.net [66.92.2.178]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3P5hC426529 for ; Thu, 25 Apr 2002 01:43:13 -0400 (EDT) Received: from wgops.com ([10.1.2.207]) by free.wgops.com (8.11.3/8.11.3) with ESMTP id g3P5hBR43020; Wed, 24 Apr 2002 22:43:11 -0700 (PDT) (envelope-from mloftis@wgops.com) Message-ID: <3CC7976F.7070407@wgops.com> Date: Wed, 24 Apr 2002 22:43:11 -0700 From: Michael Loftis User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:0.9.4.1) Gecko/20020314 Netscape6/6.2.2 X-Accept-Language: en-us MIME-Version: 1.0 To: Tom Lane cc: Curt Sampson , Bruce Momjian , PostgreSQL-development Subject: Re: [HACKERS] Sequential Scan Read-Ahead References: <12342.1019705420@sss.pgh.pa.us> Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Status: OR Tom Lane wrote: >Curt Sampson writes: > >>Grabbing bigger chunks is always optimal, AFICT, if they're not >>*too* big and you use the data. A single 64K read takes very little >>longer than a single 8K read. >> > >Proof? > I contend this statement. It's optimal to a point. I know that my system settles into it's best read-speeds @ 32K or 64K chunks. 8K chunks are far below optimal for my system. Most systems I work on do far better at 16K than at 8K, and most don't see any degradation when going to 32K chunks. (this is across numerous OSes and configs -- results are interpretations from bonnie disk i/o marks). Depending on what you're doing it is more efficiend to read bigger blocks up to a point. If you're multi-thread or reading in non-blocking mode, take as big a chunk as you can handle or are ready to process in quick order. If you're picking up a bunch of little chunks here and there and know oyu're not using them again then choose a size that will hopeuflly cause some of the reads to overlap, failing that, pick the smallest usable read size. The OS can never do that stuff for you. From cjs@cynic.net Thu Apr 25 03:29:05 2002 Return-path: Received: from angelic.cynic.net ([202.232.117.21]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3P7T3404027 for ; Thu, 25 Apr 2002 03:29:03 -0400 (EDT) Received: from localhost (localhost [127.0.0.1]) by angelic.cynic.net (Postfix) with ESMTP id 1C44E870E; Thu, 25 Apr 2002 16:28:51 +0900 (JST) Date: Thu, 25 Apr 2002 16:28:51 +0900 (JST) From: Curt Sampson To: Tom Lane cc: Bruce Momjian , PostgreSQL-development Subject: Re: [HACKERS] Sequential Scan Read-Ahead In-Reply-To: <12342.1019705420@sss.pgh.pa.us> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Status: OR On Wed, 24 Apr 2002, Tom Lane wrote: > Curt Sampson writes: > > Grabbing bigger chunks is always optimal, AFICT, if they're not > > *too* big and you use the data. A single 64K read takes very little > > longer than a single 8K read. > > Proof? Well, there are various sorts of "proof" for this assertion. What sort do you want? Here's a few samples; if you're looking for something different to satisfy you, let's discuss it. 1. Theoretical proof: two components of the delay in retrieving a block from disk are the disk arm movement and the wait for the right block to rotate under the head. When retrieving, say, eight adjacent blocks, these will be spread across no more than two cylinders (with luck, only one). The worst case access time for a single block is the disk arm movement plus the full rotational wait; this is the same as the worst case for eight blocks if they're all on one cylinder. If they're not on one cylinder, they're still on adjacent cylinders, requiring a very short seek. 2. Proof by others using it: SQL server uses 64K reads when doing table scans, as they say that their research indicates that the major limitation is usually the number of I/O requests, not the I/O capacity of the disk. BSD's explicitly separates the optimum allocation size for storage (1K fragments) and optimum read size (8K blocks) because they found performance to be much better when a larger size block was read. Most file system vendors, too, do read-ahead for this very reason. 3. Proof by testing. I wrote a little ruby program to seek to a random point in the first 2 GB of my raw disk partition and read 1-8 8K blocks of data. (This was done as one I/O request.) (Using the raw disk partition I avoid any filesystem buffering.) Here are typical results: 125 reads of 16x8K blocks: 1.9 sec, 66.04 req/sec. 15.1 ms/req, 0.946 ms/block 250 reads of 8x8K blocks: 1.9 sec, 132.3 req/sec. 7.56 ms/req, 0.945 ms/block 500 reads of 4x8K blocks: 2.5 sec, 199 req/sec. 5.03 ms/req, 1.26 ms/block 1000 reads of 2x8K blocks: 3.8 sec, 261.6 req/sec. 3.82 ms/req, 1.91 ms/block 2000 reads of 1x8K blocks: 6.4 sec, 310.4 req/sec. 3.22 ms/req, 3.22 ms/block The ratios of data retrieval speed per read for groups of adjacent 8K blocks, assuming a single 8K block reads in 1 time unit, are: 1 block 1.00 2 blocks 1.18 4 blocks 1.56 8 blocks 2.34 16 blocks 4.68 At less than 20% more expensive, certainly two-block read requests could be considered to cost "very little more" than one-block read requests. Even four-block read requests are only half-again as expensive. And if you know you're really going to be using the data, read in 8 block chunks and your cost per block (in terms of time) drops to less than a third of the cost of single-block reads. Let me put paid to comments about multiple simultaneous readers making this invalid. Here's a typical result I get with four instances of the program running simultaneously: 125 reads of 16x8K blocks: 4.4 sec, 28.21 req/sec. 35.4 ms/req, 2.22 ms/block 250 reads of 8x8K blocks: 3.9 sec, 64.88 req/sec. 15.4 ms/req, 1.93 ms/block 500 reads of 4x8K blocks: 5.8 sec, 86.52 req/sec. 11.6 ms/req, 2.89 ms/block 1000 reads of 2x8K blocks: 10 sec, 100.2 req/sec. 9.98 ms/req, 4.99 ms/block 2000 reads of 1x8K blocks: 18 sec, 110 req/sec. 9.09 ms/req, 9.09 ms/block Here's the ratio table again, with another column comparing the aggregate number of requests per second for one process and four processes: 1 block 1.00 310 : 440 2 blocks 1.10 262 : 401 4 blocks 1.28 199 : 346 8 blocks 1.69 132 : 260 16 blocks 3.89 66 : 113 Note that, here the relative increase in performance for increasing sizes of reads is even *better* until we get past 64K chunks. The overall throughput is better, of course, because with more requests per second coming in, the disk seek ordering code has more to work with and the average seek time spent seeking vs. reading will be reduced. You know, this is not rocket science; I'm sure there must be papers all over the place about this. If anybody still disagrees that it's a good thing to read chunks up to 64K or so when the blocks are adjacent and you know you'll need the data, I'd like to see some tangible evidence to support that. cjs -- Curt Sampson +81 90 7737 2974 http://www.netbsd.org Don't you know, in this new Dark Age, we're all light. --XTC From cjs@cynic.net Thu Apr 25 03:55:59 2002 Return-path: Received: from angelic.cynic.net ([202.232.117.21]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3P7tv405489 for ; Thu, 25 Apr 2002 03:55:57 -0400 (EDT) Received: from localhost (localhost [127.0.0.1]) by angelic.cynic.net (Postfix) with ESMTP id 188EC870E; Thu, 25 Apr 2002 16:55:51 +0900 (JST) Date: Thu, 25 Apr 2002 16:55:50 +0900 (JST) From: Curt Sampson To: Bruce Momjian cc: PostgreSQL-development Subject: Re: [HACKERS] Sequential Scan Read-Ahead In-Reply-To: <200204250404.g3P44OI19061@candle.pha.pa.us> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Status: OR On Thu, 25 Apr 2002, Bruce Momjian wrote: > Well, we are guilty of trying to push as much as possible on to other > software. We do this for portability reasons, and because we think our > time is best spent dealing with db issues, not issues then can be deal > with by other existing software, as long as the software is decent. That's fine. I think that's a perfectly fair thing to do. It was just the wording (i.e., "it's this other software's fault that blah de blah") that got to me. To say, "We don't do readahead becase most OSes supply it, and we feel that other things would help more to improve performance," is fine by me. Or even, "Well, nobody feels like doing it. You want it, do it yourself," I have no problem with. > Sure, that is certainly true. However, it is hard to know what the > future will hold even if we had perfect knowledge of what was happening > in the kernel. We don't know who else is going to start doing I/O once > our I/O starts. We may have a better idea with kernel knowledge, but we > still don't know 100% what will be cached. Well, we do if we use raw devices and do our own caching, using pages that are pinned in RAM. That was sort of what I was aiming at for the long run. > We have free-behind on our list. Uh...can't do it, if you're relying on the OS to do the buffering. How do you tell the OS that you're no longer going to use a page? > I think LRU-K will do this quite well > and be a nice general solution for more than just sequential scans. LRU-K sounds like a great idea to me, as does putting pages read for a table scan at the LRU end of the cache, rather than the MRU (assuming we do something to ensure that they stay in cache until read once, at any rate). But again, great for your own cache, but doesn't work with the OS cache. And I'm a bit scared to crank up too high the amount of memory I give Postgres, lest the OS try to too aggressively buffer all that I/O in what memory remains to it, and start blowing programs (like maybe the backend binary itself) out of RAM. But maybe this isn't typically a problem; I don't know. > There may be validity in this. It is easy to do (I think) and could be > a win. It didn't look to difficult to me, when I looked at the code, and you can see what kind of win it is from the response I just made to Tom. > > 1. It is *not* true that you have no idea where data is when > > using a storage array or other similar system. While you > > certainly ought not worry about things such as head positions > > and so on, it's been a given for a long, long time that two > > blocks that have close index numbers are going to be close > > together in physical storage. > > SCSI drivers, for example, are pretty smart. Not sure we can take > advantage of that from user-land I/O. Looking at the NetBSD ones, I don't see what they're doing that's so smart. (Aside from some awfully clever workarounds for stupid hardware limitations that would otherwise kill performance.) What sorts of "smart" are you referring to? > Yes, but we are seeing some db's moving away from raw I/O. Such as whom? And are you certain that they're moving to using the OS buffer cache, too? MS SQL server, for example, uses the filesystem, but turns off all buffering on those files. > Our performance numbers beat most of the big db's already, so we must > be doing something right. Really? Do the performance numbers for simple, bulk operations (imports, exports, table scans) beat the others handily? My intuition says not, but I'll happily be convinced otherwise. > Yes, but do we spend our time doing that. Is the payoff worth it, vs. > working on other features. Sure it would be great to have all these > fancy things, but is this where our time should be spent, considering > other items on the TODO list? I agree that these things need to be assesed. > Jumping in and doing the I/O ourselves is a big undertaking, and looking > at our TODO list, I am not sure if it is worth it right now. Right. I'm not trying to say this is a critical priority, I'm just trying to determine what we do right now, what we could do, and the potential performance increase that would give us. cjs -- Curt Sampson +81 90 7737 2974 http://www.netbsd.org Don't you know, in this new Dark Age, we're all light. --XTC From cjs@cynic.net Thu Apr 25 05:19:11 2002 Return-path: Received: from angelic.cynic.net ([202.232.117.21]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3P9J9412878 for ; Thu, 25 Apr 2002 05:19:10 -0400 (EDT) Received: from localhost (localhost [127.0.0.1]) by angelic.cynic.net (Postfix) with ESMTP id 50386870E; Thu, 25 Apr 2002 18:19:03 +0900 (JST) Date: Thu, 25 Apr 2002 18:19:02 +0900 (JST) From: Curt Sampson To: Tom Lane cc: Bruce Momjian , PostgreSQL-development Subject: Re: [HACKERS] Sequential Scan Read-Ahead In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Status: OR On Thu, 25 Apr 2002, Curt Sampson wrote: > Here's the ratio table again, with another column comparing the > aggregate number of requests per second for one process and four > processes: > Just for interest, I ran this again with 20 processes working simultaneously. I did six runs at each blockread size and summed the tps for each process to find the aggregate number of reads per second during the test. I dropped the higest and the lowest ones, and averaged the rest. Here's the new table: 1 proc 4 procs 20 procs 1 block 310 440 260 2 blocks 262 401 481 4 blocks 199 346 354 8 blocks 132 260 250 16 blocks 66 113 116 I'm not sure at all why performance gets so much *worse* with a lot of contention on the 1K reads. This could have something to with NetBSD, or its buffer cache, or my laptop's crappy little disk drive.... Or maybe I'm just running out of CPU. cjs -- Curt Sampson +81 90 7737 2974 http://www.netbsd.org Don't you know, in this new Dark Age, we're all light. --XTC From tgl@sss.pgh.pa.us Thu Apr 25 09:54:35 2002 Return-path: Received: from sss.pgh.pa.us (root@[192.204.191.242]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3PDsY407038 for ; Thu, 25 Apr 2002 09:54:34 -0400 (EDT) Received: from sss2.sss.pgh.pa.us (tgl@localhost [127.0.0.1]) by sss.pgh.pa.us (8.11.4/8.11.4) with ESMTP id g3PDsXF25059; Thu, 25 Apr 2002 09:54:33 -0400 (EDT) To: Curt Sampson cc: Bruce Momjian , PostgreSQL-development Subject: Re: [HACKERS] Sequential Scan Read-Ahead In-Reply-To: References: Comments: In-reply-to Curt Sampson message dated "Thu, 25 Apr 2002 16:28:51 +0900" Date: Thu, 25 Apr 2002 09:54:32 -0400 Message-ID: <25056.1019742872@sss.pgh.pa.us> From: Tom Lane Status: OR Curt Sampson writes: > 1. Theoretical proof: two components of the delay in retrieving a > block from disk are the disk arm movement and the wait for the > right block to rotate under the head. > When retrieving, say, eight adjacent blocks, these will be spread > across no more than two cylinders (with luck, only one). Weren't you contending earlier that with modern disk mechs you really have no idea where the data is? You're asserting as an article of faith that the OS has been able to place the file's data blocks optimally --- or at least well enough to avoid unnecessary seeks. But just a few days ago I was getting told that random_page_cost was BS because there could be no such placement. I'm getting a tad tired of sweeping generalizations offered without proof, especially when they conflict. > 3. Proof by testing. I wrote a little ruby program to seek to a > random point in the first 2 GB of my raw disk partition and read > 1-8 8K blocks of data. (This was done as one I/O request.) (Using > the raw disk partition I avoid any filesystem buffering.) And also ensure that you aren't testing the point at issue. The point at issue is that *in the presence of kernel read-ahead* it's quite unclear that there's any benefit to a larger request size. Ideally the kernel will have the next block ready for you when you ask, no matter what the request is. There's been some talk of using the AIO interface (where available) to "encourage" the kernel to do read-ahead. I don't foresee us writing our own substitute filesystem to make this happen, however. Oracle may have the manpower for that sort of boondoggle, but we don't... regards, tom lane From pgsql-hackers-owner+M22053@postgresql.org Thu Apr 25 20:45:42 2002 Return-path: Received: from postgresql.org (postgresql.org [64.49.215.8]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3Q0jg405210 for ; Thu, 25 Apr 2002 20:45:42 -0400 (EDT) Received: from postgresql.org (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with SMTP id 17CE6476270; Thu, 25 Apr 2002 20:45:38 -0400 (EDT) Received: from doppelbock.patentinvestor.com (ip146.usw5.rb1.bel.nwlink.com [209.20.249.146]) by postgresql.org (Postfix) with ESMTP id 257DC47591C for ; Thu, 25 Apr 2002 20:45:25 -0400 (EDT) Received: (from kaf@localhost) by doppelbock.patentinvestor.com (8.11.6/8.11.2) id g3Q0erX14397; Thu, 25 Apr 2002 17:40:53 -0700 From: Kyle MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <15560.41493.529847.635632@doppelbock.patentinvestor.com> Date: Thu, 25 Apr 2002 17:40:53 -0700 To: PostgreSQL-development Subject: Re: [HACKERS] Sequential Scan Read-Ahead In-Reply-To: <25056.1019742872@sss.pgh.pa.us> References: <25056.1019742872@sss.pgh.pa.us> X-Mailer: VM 6.95 under 21.1 (patch 14) "Cuyahoga Valley" XEmacs Lucid Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org Status: ORr Tom Lane wrote: > ... > Curt Sampson writes: > > 3. Proof by testing. I wrote a little ruby program to seek to a > > random point in the first 2 GB of my raw disk partition and read > > 1-8 8K blocks of data. (This was done as one I/O request.) (Using > > the raw disk partition I avoid any filesystem buffering.) > > And also ensure that you aren't testing the point at issue. > The point at issue is that *in the presence of kernel read-ahead* > it's quite unclear that there's any benefit to a larger request size. > Ideally the kernel will have the next block ready for you when you > ask, no matter what the request is. > ... I have to agree with Tom. I think the numbers below show that with kernel read-ahead, block size isn't an issue. The big_file1 file used below is 2.0 gig of random data, and the machine has 512 mb of main memory. This ensures that we're not just getting cached data. foreach i (4k 8k 16k 32k 64k 128k) echo $i time dd bs=$i if=big_file1 of=/dev/null end and the results: bs user kernel elapsed 4k: 0.260 7.740 1:27.25 8k: 0.210 8.060 1:30.48 16k: 0.090 7.790 1:30.88 32k: 0.060 8.090 1:32.75 64k: 0.030 8.190 1:29.11 128k: 0.070 9.830 1:28.74 so with kernel read-ahead, we have basically the same elapsed (wall time) regardless of block size. Sure, user time drops to a low at 64k blocksize, but kernel time is increasing. You could argue that this is a contrived example, no other I/O is being done. Well I created a second 2.0g file (big_file2) and did two simultaneous reads from the same disk. Sure performance went to hell but it shows blocksize is still irrelevant in a multi I/O environment with sequential read-ahead. foreach i ( 4k 8k 16k 32k 64k 128k ) echo $i time dd bs=$i if=big_file1 of=/dev/null & time dd bs=$i if=big_file2 of=/dev/null & wait end bs user kernel elapsed 4k: 0.480 8.290 6:34.13 bigfile1 0.320 8.730 6:34.33 bigfile2 8k: 0.250 7.580 6:31.75 0.180 8.450 6:31.88 16k: 0.150 8.390 6:32.47 0.100 7.900 6:32.55 32k: 0.190 8.460 6:24.72 0.060 8.410 6:24.73 64k: 0.060 9.350 6:25.05 0.150 9.240 6:25.13 128k: 0.090 10.610 6:33.14 0.110 11.320 6:33.31 the differences in read times are basically in the mud. Blocksize just doesn't matter much with the kernel doing readahead. -Kyle ---------------------------(end of broadcast)--------------------------- TIP 6: Have you searched our list archives? http://archives.postgresql.org From pgsql-hackers-owner+M22055@postgresql.org Thu Apr 25 22:19:07 2002 Return-path: Received: from postgresql.org (postgresql.org [64.49.215.8]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3Q2J7411254 for ; Thu, 25 Apr 2002 22:19:07 -0400 (EDT) Received: from postgresql.org (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with SMTP id F3924476208; Thu, 25 Apr 2002 22:19:02 -0400 (EDT) Received: from candle.pha.pa.us (216-55-132-35.dsl.san-diego.abac.net [216.55.132.35]) by postgresql.org (Postfix) with ESMTP id 6741D474E71 for ; Thu, 25 Apr 2002 22:18:50 -0400 (EDT) Received: (from pgman@localhost) by candle.pha.pa.us (8.11.6/8.10.1) id g3Q2Ili11246; Thu, 25 Apr 2002 22:18:47 -0400 (EDT) From: Bruce Momjian Message-ID: <200204260218.g3Q2Ili11246@candle.pha.pa.us> Subject: Re: [HACKERS] Sequential Scan Read-Ahead In-Reply-To: <15560.41493.529847.635632@doppelbock.patentinvestor.com> To: Kyle Date: Thu, 25 Apr 2002 22:18:47 -0400 (EDT) cc: PostgreSQL-development X-Mailer: ELM [version 2.4ME+ PL97 (25)] MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII Precedence: bulk Sender: pgsql-hackers-owner@postgresql.org Status: OR Nice test. Would you test simultaneous 'dd' on the same file, perhaps with a slight delay between to the two so they don't read each other's blocks? seek() in the file will turn off read-ahead in most OS's. I am not saying this is a major issue for PostgreSQL but the numbers would be interesting. --------------------------------------------------------------------------- Kyle wrote: > Tom Lane wrote: > > ... > > Curt Sampson writes: > > > 3. Proof by testing. I wrote a little ruby program to seek to a > > > random point in the first 2 GB of my raw disk partition and read > > > 1-8 8K blocks of data. (This was done as one I/O request.) (Using > > > the raw disk partition I avoid any filesystem buffering.) > > > > And also ensure that you aren't testing the point at issue. > > The point at issue is that *in the presence of kernel read-ahead* > > it's quite unclear that there's any benefit to a larger request size. > > Ideally the kernel will have the next block ready for you when you > > ask, no matter what the request is. > > ... > > I have to agree with Tom. I think the numbers below show that with > kernel read-ahead, block size isn't an issue. > > The big_file1 file used below is 2.0 gig of random data, and the > machine has 512 mb of main memory. This ensures that we're not > just getting cached data. > > foreach i (4k 8k 16k 32k 64k 128k) > echo $i > time dd bs=$i if=big_file1 of=/dev/null > end > > and the results: > > bs user kernel elapsed > 4k: 0.260 7.740 1:27.25 > 8k: 0.210 8.060 1:30.48 > 16k: 0.090 7.790 1:30.88 > 32k: 0.060 8.090 1:32.75 > 64k: 0.030 8.190 1:29.11 > 128k: 0.070 9.830 1:28.74 > > so with kernel read-ahead, we have basically the same elapsed (wall > time) regardless of block size. Sure, user time drops to a low at 64k > blocksize, but kernel time is increasing. > > > You could argue that this is a contrived example, no other I/O is > being done. Well I created a second 2.0g file (big_file2) and did two > simultaneous reads from the same disk. Sure performance went to hell > but it shows blocksize is still irrelevant in a multi I/O environment > with sequential read-ahead. > > foreach i ( 4k 8k 16k 32k 64k 128k ) > echo $i > time dd bs=$i if=big_file1 of=/dev/null & > time dd bs=$i if=big_file2 of=/dev/null & > wait > end > > bs user kernel elapsed > 4k: 0.480 8.290 6:34.13 bigfile1 > 0.320 8.730 6:34.33 bigfile2 > 8k: 0.250 7.580 6:31.75 > 0.180 8.450 6:31.88 > 16k: 0.150 8.390 6:32.47 > 0.100 7.900 6:32.55 > 32k: 0.190 8.460 6:24.72 > 0.060 8.410 6:24.73 > 64k: 0.060 9.350 6:25.05 > 0.150 9.240 6:25.13 > 128k: 0.090 10.610 6:33.14 > 0.110 11.320 6:33.31 > > > the differences in read times are basically in the mud. Blocksize > just doesn't matter much with the kernel doing readahead. > > -Kyle > > ---------------------------(end of broadcast)--------------------------- > TIP 6: Have you searched our list archives? > > http://archives.postgresql.org > -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026 ---------------------------(end of broadcast)--------------------------- TIP 6: Have you searched our list archives? http://archives.postgresql.org From cjs@cynic.net Thu Apr 25 22:27:23 2002 Return-path: Received: from angelic.cynic.net ([202.232.117.21]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3Q2RL411868 for ; Thu, 25 Apr 2002 22:27:22 -0400 (EDT) Received: from localhost (localhost [127.0.0.1]) by angelic.cynic.net (Postfix) with ESMTP id AF60C870E; Fri, 26 Apr 2002 11:27:17 +0900 (JST) Date: Fri, 26 Apr 2002 11:27:17 +0900 (JST) From: Curt Sampson To: Tom Lane cc: Bruce Momjian , PostgreSQL-development Subject: Re: [HACKERS] Sequential Scan Read-Ahead In-Reply-To: <25056.1019742872@sss.pgh.pa.us> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Status: OR On Thu, 25 Apr 2002, Tom Lane wrote: > Curt Sampson writes: > > 1. Theoretical proof: two components of the delay in retrieving a > > block from disk are the disk arm movement and the wait for the > > right block to rotate under the head. > > > When retrieving, say, eight adjacent blocks, these will be spread > > across no more than two cylinders (with luck, only one). > > Weren't you contending earlier that with modern disk mechs you really > have no idea where the data is? No, that was someone else. I contend that with pretty much any large-scale storage mechanism (i.e., anything beyond ramdisks), you will find that accessing two adjacent blocks is almost always 1) close to as fast as accessing just the one, and 2) much, much faster than accessing two blocks that are relatively far apart. There will be the odd case where the two adjacent blocks are physically far apart, but this is rare. If this idea doesn't hold true, the whole idea that sequential reads are faster than random reads falls apart, and the optimizer shouldn't even have the option to make random reads cost more, much less have it set to four rather than one (or whatever it's set to). > You're asserting as an article of > faith that the OS has been able to place the file's data blocks > optimally --- or at least well enough to avoid unnecessary seeks. So are you, in the optimizer. But that's all right; the OS often can and does do this placement; the FFS filesystem is explicitly designed to do this sort of thing. If the filesystem isn't empty and the files grow a lot they'll be split into large fragments, but the fragments will be contiguous. > But just a few days ago I was getting told that random_page_cost > was BS because there could be no such placement. I've been arguing against that point as well. > And also ensure that you aren't testing the point at issue. > The point at issue is that *in the presence of kernel read-ahead* > it's quite unclear that there's any benefit to a larger request size. I will test this. cjs -- Curt Sampson +81 90 7737 2974 http://www.netbsd.org Don't you know, in this new Dark Age, we're all light. --XTC From cjs@cynic.net Wed Apr 24 23:19:23 2002 Return-path: Received: from angelic.cynic.net ([202.232.117.21]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3P3JM414917 for ; Wed, 24 Apr 2002 23:19:22 -0400 (EDT) Received: from localhost (localhost [127.0.0.1]) by angelic.cynic.net (Postfix) with ESMTP id 1F36F870E; Thu, 25 Apr 2002 12:19:14 +0900 (JST) Date: Thu, 25 Apr 2002 12:19:14 +0900 (JST) From: Curt Sampson To: Bruce Momjian cc: PostgreSQL-development Subject: Re: Sequential Scan Read-Ahead In-Reply-To: <200204250156.g3P1ufh05751@candle.pha.pa.us> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Status: OR On Wed, 24 Apr 2002, Bruce Momjian wrote: > > 1. Not all systems do readahead. > > If they don't, that isn't our problem. We expect it to be there, and if > it isn't, the vendor/kernel is at fault. It is your problem when another database kicks Postgres' ass performance-wise. And at that point, *you're* at fault. You're the one who's knowingly decided to do things inefficiently. Sorry if this sounds harsh, but this, "Oh, someone else is to blame" attitude gets me steamed. It's one thing to say, "We don't support this." That's fine; there are often good reasons for that. It's a completely different thing to say, "It's an unrelated entity's fault we don't support this." At any rate, relying on the kernel to guess how to optimise for the workload will never work as well as well as the software that knows the workload doing the optimization. The lack of support thing is no joke. Sure, lots of systems nowadays support unified buffer cache and read-ahead. But how many, besides Solaris, support free-behind, which is also very important to avoid blowing out your buffer cache when doing sequential reads? And who at all supports read-ahead for reverse scans? (Or does Postgres not do those, anyway? I can see the support is there.) And even when the facilities are there, you create problems by using them. Look at the OS buffer cache, for example. Not only do we lose efficiency by using two layers of caching, but (as people have pointed out recently on the lists), the optimizer can't even know how much or what is being cached, and thus can't make decisions based on that. > Yes, seek() in file will turn off read-ahead. Grabbing bigger chunks > would help here, but if you have two people already reading from the > same file, grabbing bigger chunks of the file may not be optimal. Grabbing bigger chunks is always optimal, AFICT, if they're not *too* big and you use the data. A single 64K read takes very little longer than a single 8K read. > > 3. Even when the read-ahead does occur, you're still doing more > > syscalls, and thus more expensive kernel/userland transitions, than > > you have to. > > I would guess the performance impact is minimal. If it were minimal, people wouldn't work so hard to build multi-level thread systems, where multiple userland threads are scheduled on top of kernel threads. However, it does depend on how much CPU your particular application is using. You may have it to spare. > http://candle.pha.pa.us/mhonarc/todo.detail/performance/msg00009.html Well, this message has some points in it that I feel are just incorrect. 1. It is *not* true that you have no idea where data is when using a storage array or other similar system. While you certainly ought not worry about things such as head positions and so on, it's been a given for a long, long time that two blocks that have close index numbers are going to be close together in physical storage. 2. Raw devices are quite standard across Unix systems (except in the unfortunate case of Linux, which I think has been remedied, hasn't it?). They're very portable, and have just as well--if not better--defined write semantics as a filesystem. 3. My observations of OS performance tuning over the past six or eight years contradict the statement, "There's a considerable cost in complexity and code in using "raw" storage too, and it's not a one off cost: as the technologies change, the "fast" way to do things will change and the code will have to be updated to match." While optimizations have been removed over the years the basic optimizations (order reads by block number, do larger reads rather than smaller, cache the data) have remained unchanged for a long, long time. 4. "Better to leave this to the OS vendor where possible, and take advantage of the tuning they do." Well, sorry guys, but have a look at the tuning they do. It hasn't changed in years, except to remove now-unnecessary complexity realated to really, really old and slow disk devices, and to add a few thing that guess workload but still do a worse job than if the workload generator just did its own optimisations in the first place. > http://candle.pha.pa.us/mhonarc/todo.detail/optimizer/msg00011.html Well, this one, with statements like "Postgres does have control over its buffer cache," I don't know what to say. You can interpret the statement however you like, but in the end Postgres very little control at all over how data is moved between memory and disk. BTW, please don't take me as saying that all control over physical IO should be done by Postgres. I just think that Posgres could do a better job of managing data transfer between disk and memory than the OS can. The rest of the things (using raw paritions, read-ahead, free-behind, etc.) just drop out of that one idea. cjs -- Curt Sampson +81 90 7737 2974 http://www.netbsd.org Don't you know, in this new Dark Age, we're all light. --XTC From kaf@nwlink.com Fri Apr 26 14:22:39 2002 Return-path: Received: from doppelbock.patentinvestor.com (ip146.usw5.rb1.bel.nwlink.com [209.20.249.146]) by candle.pha.pa.us (8.11.6/8.10.1) with ESMTP id g3QIMc400783 for ; Fri, 26 Apr 2002 14:22:38 -0400 (EDT) Received: (from kaf@localhost) by doppelbock.patentinvestor.com (8.11.6/8.11.2) id g3QII0l16824; Fri, 26 Apr 2002 11:18:00 -0700 From: Kyle MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <15561.39384.296503.501888@doppelbock.patentinvestor.com> Date: Fri, 26 Apr 2002 11:18:00 -0700 To: Bruce Momjian Subject: Re: [HACKERS] Sequential Scan Read-Ahead In-Reply-To: <200204261444.g3QEiFh11090@candle.pha.pa.us> References: <15561.26116.817541.950416@doppelbock.patentinvestor.com> <200204261444.g3QEiFh11090@candle.pha.pa.us> X-Mailer: VM 6.95 under 21.1 (patch 14) "Cuyahoga Valley" XEmacs Lucid Status: ORr Hey Bruce, I'll forward this to the list if you think they'd benefit from it. I'm not sure it says anything about read-ahead, I think this is more a kernel caching issue. But I've been known to be wrong in the past. Anyway... the test: foreach i (5 15 20 25 30 ) echo $i time dd bs=8k if=big_file1 of=/dev/null & sleep $i time dd bs=8k if=big_file1 of=/dev/null & wait end I did a couple more runs in the low range since their is a drastic jump in elapsed (wall clock) time after doing a 6 second sleep: first process second process sleep user kernel elapsed user kernel elapsed 0 sec 0.200 7.980 1:26.57 0.240 7.720 1:26.56 3 sec 0.260 7.600 1:25.71 0.260 8.100 1:22.60 5 sec 0.160 7.890 1:26.04 0.220 8.180 1:21.04 6 sec 0.220 8.070 1:19.59 0.230 7.620 1:25.69 7 sec 0.210 9.270 1:57.92 0.100 8.750 1:50.76 8 sec 0.240 8.060 4:47.47 0.300 7.800 4:40.40 15 sec 0.200 8.500 4:51.11 0.180 7.280 4:44.36 20 sec 0.160 8.040 4:40.72 0.240 7.790 4:37.24 25 sec 0.170 8.150 4:37.58 0.140 8.200 4:33.08 30 sec 0.200 7.390 4:37.01 0.230 8.220 4:31.83 with a sleep of > 6 seconds, either the second process isn't getting cached data or readahead is being turned off. I'd guess the former, I don't see why read-ahead would be turned off since they're both doing sequential operations. Although with 512mb of memory and the disk reading at about 22 mb/sec, maybe we're not hitting the cache. I'd guess at least ~400 megs of kernel cache is being used for buffering this 2 gig file. free(1) reports: % free total used free shared buffers cached Mem: 512924 508576 4348 0 2640 477960 -/+ buffers/cache: 27976 484948 Swap: 527152 15864 511288 so shouldn't we be getting cached data even with a sleep of up to about (400/22) 18 seconds...? Maybe I'm just in the dark on what's really happening. I should point out that this is linux 2.4.18. Bruce Momjian wrote: > > I am trying to illustrate how kernel read-ahead could be turned off in > certain cases. > > --------------------------------------------------------------------------- > > Kyle wrote: > > What are you trying to test, the kernel's cache vs disk speed? > > > > > > Bruce Momjian wrote: > > > > > > Nice test. Would you test simultaneous 'dd' on the same file, perhaps > > > with a slight delay between to the two so they don't read each other's > > > blocks? > > > > > > seek() in the file will turn off read-ahead in most OS's. I am not > > > saying this is a major issue for PostgreSQL but the numbers would be > > > interesting.