Comments on: MySQL Users Conference – Innodb http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/ Everything about MySQL Performance Sat, 07 Nov 2009 18:35:44 -0800 http://wordpress.org/?v=2.8.4 hourly 1 By: Mark Callaghan http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/comment-page-1/#comment-122316 Mark Callaghan Fri, 11 May 2007 18:38:25 +0000 http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/#comment-122316 The big patch for MySQL 5 and the semi-sync only patch for MySQL 5 are at http://google-mysql-tools.googlecode.com/svn/trunk/mysql-patches. Real async IO has one advantage over background IO threads that I am curious about. With async IO, if an extent is to be prefetched, then 1 or more async IO requests can be used. While a large IO request can be submitted for some of the pages, the prefix or suffix of the extent (depending on scan direction) should use one request per page so that the page is available before the rest of the pages. Will we ever see the work done by the DIKU researchers show up in InnoDB? Until then, or until async IO is proven to work with InnoDB, extra background IO threads might be the best alternative. The big patch for MySQL 5 and the semi-sync only patch for MySQL 5 are at http://google-mysql-tools.googlecode.com/svn/trunk/mysql-patches.

Real async IO has one advantage over background IO threads that I am curious about. With async IO, if an extent is to be prefetched, then 1 or more async IO requests can be used. While a large IO request can be submitted for some of the pages, the prefix or suffix of the extent (depending on scan direction) should use one request per page so that the page is available before the rest of the pages. Will we ever see the work done by the DIKU researchers show up in InnoDB? Until then, or until async IO is proven to work with InnoDB, extra background IO threads might be the best alternative.

]]> By: Talking About MySQL « François Schiettecatte’s Blog http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/comment-page-1/#comment-121224 Talking About MySQL « François Schiettecatte’s Blog Wed, 09 May 2007 21:30:50 +0000 http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/#comment-121224 [...] About MySQL While I am talking about database and scaling, Peter Zaitsev has some notes on new Innodb features being discussed at the recent MySQL Users [...] [...] About MySQL While I am talking about database and scaling, Peter Zaitsev has some notes on new Innodb features being discussed at the recent MySQL Users [...]

]]> By: peter http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/comment-page-1/#comment-121117 peter Wed, 09 May 2007 18:10:00 +0000 http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/#comment-121117 Thank you Mark, That is great. I see you've got a lot of interest in Google patches to port them to MySQL 5.0 quickly :) I've checked MySQL 5.0 patches page it seems to only list one patch which is for semi-synchronous replication. Where is other big one. Regarding number of IO threads... well if Asynchronous IO is available and indeed well utilized, ie thread would never block with single outstanding IO it is good enough. Thank you Mark,

That is great. I see you’ve got a lot of interest in Google patches to port them to MySQL 5.0 quickly :)

I’ve checked MySQL 5.0 patches page it seems to only list one patch which is for semi-synchronous replication. Where is other big one.

Regarding number of IO threads… well if Asynchronous IO is available and indeed well utilized, ie thread would never block with single outstanding IO it is good enough.

]]> By: Mark Callaghan http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/comment-page-1/#comment-121093 Mark Callaghan Wed, 09 May 2007 17:49:50 +0000 http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/#comment-121093 I think you have overstated what was provided in the Google patch, but we don't mind the nice words. The patches from Google were first available for MySQL 4.0.26. They are now available for 4.0.26 and 5.0.37 -- see http://code.google.com/p/google-mysql-tools to get them. There are two patches for 5.0.37, one that only has the changes for semi-sync replication and another that has everything. The changes for 4.0 have gotten more use than the changes for 5.0 -- I have yet to do IO performance testing for 5.0. In theory, the changes for more IO threads in InnoDB are not needed, as InnoDB has support for using async IO, but this is disabled because async IO didn't work too well on Linux in the past. Now that Oracle runs on Linux and uses async IO, it should work well. I think you have overstated what was provided in the Google patch, but we don’t mind the nice words. The patches from Google were first available for MySQL 4.0.26. They are now available for 4.0.26 and 5.0.37 — see http://code.google.com/p/google-mysql-tools to get them. There are two patches for 5.0.37, one that only has the changes for semi-sync replication and another that has everything. The changes for 4.0 have gotten more use than the changes for 5.0 — I have yet to do IO performance testing for 5.0. In theory, the changes for more IO threads in InnoDB are not needed, as InnoDB has support for using async IO, but this is disabled because async IO didn’t work too well on Linux in the past. Now that Oracle runs on Linux and uses async IO, it should work well.

]]> By: peter http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/comment-page-1/#comment-120871 peter Wed, 09 May 2007 11:50:39 +0000 http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/#comment-120871 I actually meant 4.0 I actually meant 4.0

]]> By: guillaume http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/comment-page-1/#comment-120836 guillaume Wed, 09 May 2007 11:25:01 +0000 http://www.mysqlperformanceblog.com/2007/05/09/mysql-users-conference-innodb/#comment-120836 Hello Peter, Thanks for the sum-up on upcoming InnoDB changes! Any performance improvements will be welcome :) About Google patches, you wrote "These are for MySQL 5.0 at this point" I suppose you meant 4.1, right? Hello Peter,

Thanks for the sum-up on upcoming InnoDB changes! Any performance improvements will be welcome :)

About Google patches, you wrote “These are for MySQL 5.0 at this point”
I suppose you meant 4.1, right?

]]>