Comments on: The MySQL optimizer, the OS cache, and sequential versus random I/O http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/ Everything about MySQL Performance Sat, 21 Nov 2009 05:23:57 -0800 http://wordpress.org/?v=2.8.4 hourly 1 By: Baron Schwartz http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/comment-page-1/#comment-511301 Baron Schwartz Wed, 18 Mar 2009 17:51:17 +0000 http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/#comment-511301 Oh, wow, did I really screw the math up that badly? I believe you are right. Funny that no one else caught it yet. You can assume that each random I/O has to wait for the head to move, for the disk to rotate the starting point of the read under the head (1/2 of a rotation on average) and then the disk to rotate all the data under the head. So the spindle rotation speed really dominates, and you can just estimate the random I/O off that. You can get an absolute upper bound easily, and in practice you won't get that performance. For sequential I/O, I have to think about that for a moment, and suddenly I'm not confident in my math skills. Oh, wow, did I really screw the math up that badly? I believe you are right. Funny that no one else caught it yet.

You can assume that each random I/O has to wait for the head to move, for the disk to rotate the starting point of the read under the head (1/2 of a rotation on average) and then the disk to rotate all the data under the head. So the spindle rotation speed really dominates, and you can just estimate the random I/O off that. You can get an absolute upper bound easily, and in practice you won’t get that performance.

For sequential I/O, I have to think about that for a moment, and suddenly I’m not confident in my math skills.

]]> By: Shiv http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/comment-page-1/#comment-511232 Shiv Wed, 18 Mar 2009 16:02:08 +0000 http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/#comment-511232 Baron- wonderful article. One of those articles that establishes faith that the beast can be understood. Question: I am not clear when at the very end you say "313 sequential reads (150 million rows / 117 bytes per row / 4096 bytes per read)". Should'nt the math be "(150 M rows*117 bytes per row)/block_size(=4096)= 4.2 M sequential reads". If so, is there a rule of thumb for how fast sequential reads can be done by commodity hardware ? (I am assuming that the number of 100 IOs per second refers to random I/O). Thanks, Shiv Baron- wonderful article. One of those articles that establishes faith that the beast can be understood.

Question: I am not clear when at the very end you say “313 sequential reads (150 million rows / 117 bytes per row / 4096 bytes per read)”. Should’nt the math be “(150 M rows*117 bytes per row)/block_size(=4096)= 4.2 M sequential reads”.

If so, is there a rule of thumb for how fast sequential reads can be done by commodity hardware ? (I am assuming that the number of 100 IOs per second refers to random I/O).

Thanks,
Shiv

]]> By: High-Performance Click Analysis with MySQL | MySQL Performance Blog http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/comment-page-1/#comment-421276 High-Performance Click Analysis with MySQL | MySQL Performance Blog Tue, 23 Dec 2008 03:48:32 +0000 http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/#comment-421276 [...] be fast. The star schema is essentially "I admit defeat and accept table scans as a fact of life." Table scans can be better than the alternative, if the alternatives are limited, but they're still not what you need unless you're okay with long [...] [...] be fast. The star schema is essentially “I admit defeat and accept table scans as a fact of life.” Table scans can be better than the alternative, if the alternatives are limited, but they’re still not what you need unless you’re okay with long [...]

]]> By: dbnewz » Blog Archive » Cardinalité, sélectivité et distributivité d’un index MySQL : quel impact sur le plan d’exécution ? http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/comment-page-1/#comment-351928 dbnewz » Blog Archive » Cardinalité, sélectivité et distributivité d’un index MySQL : quel impact sur le plan d’exécution ? Thu, 04 Sep 2008 22:38:50 +0000 http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/#comment-351928 [...] Une excellente illustration des difficultés parfois rencontrées par l’optimiseur, un coup de pouce du DBA permet alors de réduire la requête de 2.5 jours à 1h ! [...] [...] Une excellente illustration des difficultés parfois rencontrées par l’optimiseur, un coup de pouce du DBA permet alors de réduire la requête de 2.5 jours à 1h ! [...]

]]> By: Yasufumi Kinoshita http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/comment-page-1/#comment-299242 Yasufumi Kinoshita Thu, 15 May 2008 05:55:40 +0000 http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/#comment-299242 I have tried to fix 5.0 optimizer. (But I have focused on InnoDB) My patch may solve this type of problems. http://bugs.mysql.com/bug.php?id=36681 Regards. I have tried to fix 5.0 optimizer. (But I have focused on InnoDB)
My patch may solve this type of problems.

http://bugs.mysql.com/bug.php?id=36681

Regards.

]]> By: Baron Schwartz http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/comment-page-1/#comment-298520 Baron Schwartz Tue, 13 May 2008 22:00:27 +0000 http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/#comment-298520 Triffids, I don't know much about the multi-range read feature. (I say that about lots of things until I have studied it a lot). It's unfinished, and I would bet the final implementation will be somewhat different than it is now. Triffids, I don’t know much about the multi-range read feature. (I say that about lots of things until I have studied it a lot). It’s unfinished, and I would bet the final implementation will be somewhat different than it is now.

]]> By: Triffids http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/comment-page-1/#comment-298489 Triffids Tue, 13 May 2008 18:53:41 +0000 http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/#comment-298489 Cool feature ! and how MySQL optimizer chose random read (and read some blocks from cache) or use MRR (all blocks from HDD) ? Cool feature ! and how MySQL optimizer chose random read (and read some blocks from cache) or use MRR (all blocks from HDD) ?

]]> By: Baron Schwartz http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/comment-page-1/#comment-298472 Baron Schwartz Tue, 13 May 2008 16:58:48 +0000 http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/#comment-298472 Triffids, see http://www.mysqlperformanceblog.com/2008/03/25/mysql-60-vs-51-in-tpc-h-queries/ Triffids, see http://www.mysqlperformanceblog.com/2008/03/25/mysql-60-vs-51-in-tpc-h-queries/

]]> By: Triffids http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/comment-page-1/#comment-298459 Triffids Tue, 13 May 2008 16:15:55 +0000 http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/#comment-298459 And why MySQL doesn't use scattered read (multiblock read) as it do rdbms leaders ? And why MySQL doesn’t use scattered read (multiblock read) as it do rdbms leaders ?

]]> By: Webadmin rövidhírek #2 at ‹Webakadémia /› http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/comment-page-1/#comment-294116 Webadmin rövidhírek #2 at ‹Webakadémia /› Mon, 05 May 2008 07:35:57 +0000 http://www.mysqlperformanceblog.com/2008/04/28/the-mysql-optimizer-the-os-cache-and-sequential-versus-random-io/#comment-294116 [...] amikor nem árt a rásegítés. Egy konkrét optimalizálással kapcsolatos példát olvashatunk az operációs rendszer cache és a lemezműveletek kapcsán a MySQL Performance [...] [...] amikor nem árt a rásegítés. Egy konkrét optimalizálással kapcsolatos példát olvashatunk az operációs rendszer cache és a lemezműveletek kapcsán a MySQL Performance [...]

]]>