Comments on: Heikki Tuuri Innodb answers – Part I

By: mustafa vaghjipurwala

mustafa vaghjipurwala — Thu, 21 Jul 2011 15:09:50 +0000

how to display the second portion
in mysql
plzz help

By: Jason Plant

Jason Plant — Wed, 18 Nov 2009 04:46:32 +0000

christ, your english is horrible.

By: bnm

bnm — Wed, 03 Dec 2008 16:22:27 +0000

ciao
c hanno parlato d te…allora ki siamo

By: Kunal Jain

Kunal Jain — Sat, 29 Nov 2008 09:41:51 +0000

I have a QuadCore Server(4 Cores). How can i configure mysql to use all the four cores. Right now mysql process use only one core whose utilization goes upto 95%.

Thanks in advance
Kunal Jain

By: Heikki Tuuri answers to Innodb questions, Part II | MySQL Performance Blog

Heikki Tuuri answers to Innodb questions, Part II | MySQL Performance Blog — Sun, 04 Nov 2007 18:29:28 +0000

[...] I now got answers to the second portions of the questions you asked Heikki. If you have not seen the first part it can be found here. [...]

By: Pythian Group Blog » Blog Archive » Log Buffer #69: a Carnival of the Vanities for DBAs

Fri, 02 Nov 2007 17:23:19 +0000

[...] from Peter, the first part of a Q&A on InnoDB with Heikki Tuuri, CEO of Innobase and creator of InnoDB. The horse’s mouth, in other [...]

By: peter

peter — Tue, 30 Oct 2007 23:10:14 +0000

Kevin, Sergei

Indeed multiple keycaches and midpoint insertions help for MyISAM. Though I would like to see some experiments to see how well midpoint insertion is able to preserve cache from being washed away by background load.

The file cache for MyISAM is however serious issue and not only that but Disk Scheduling which in many cases gives way too much preference to large size read/writes which batch jobs may do. At least I have not found IO scheduler yet which would not behave such way.

By: Kevin Burton

Kevin Burton — Tue, 30 Oct 2007 08:27:10 +0000

Hey Sergei,

You’re right with regard to the key buffer but I was specifically thinking about the filesystem cache.

There’s now way for the filesystem cache to know what’s going on in the application layer.

I actually think there needs to be more coordination here between abstractions.

For example, if the OS is running out of memory and about to swap it could send a signal to user level applications to tell them to free up their own internal caches to return some memory to the OS.

Kevin

By: Sergei Golubchik

Sergei Golubchik — Tue, 30 Oct 2007 08:22:36 +0000

Kevin, you wrote as a comment to Q7:

“Note that most database systems like MyISAM are very vulnerable to this problem.”

MyISAM is not. At least not, as long as MySQL is concerned (filesystem caches are vulnerable, of course).
There are two protections in MyISAM. First – keycache consists of warm and hot parts. Pages are cached in the warm part, and moved to the hot part after receiving a certain number of hits. Each part has its own LRU list. That is a full index scan will only wipe out the warm part of the keycache. Pages that are used most often and that are most important to cache will stay cached in the hot cache. Read more here: http://dev.mysql.com/doc/refman/5.0/en/midpoint-insertion.html

A second protection: multiple key caches. A table may be assigned to a separate keycache, whatever you’ll do with it it will never affect cached pages of other tables. More here: http://dev.mysql.com/doc/refman/5.0/en/multiple-key-caches.html

By: Kevin Burton

Kevin Burton — Tue, 30 Oct 2007 02:26:35 +0000

Bill,

Your comment about running multiple MySQL daemons per box is interesting.

I actually blogged about this a while ago:

http://feedblog.org/2007/09/23/distributed-computing-fallacy-9/

The theory being that a quad core machine with 32G of memory and 4 disks should really be considered 4 logical machines and not one.

Also, I blogged my thoughts about this post here:

http://feedblog.org/2007/10/29/thoughts-on-innodb-internals-re-heikki-tuur/