Comments on: Why you should ignore MySQL’s key cache hit ratio

By: Akhthar

Akhthar — Tue, 25 Oct 2011 15:06:42 +0000

I would like to disagree. Although I agree that it’s a waste of time to only look at ratio and be done with Key buffer, I think it would be helpful to understand the Efficiency of Key buffer as well as compare the same with multiple servers with only one condition that the Uptime should complete a time cycle (24 hours if you are not sure about the cycle) and should be equal for all. You can still get the per day stats, but you won’t get the accurate results unless you have same Uptime for all servers.

By: Baron Schwartz

Baron Schwartz — Thu, 02 Dec 2010 18:49:08 +0000

Gabek,

Percona offers professional services for MySQL and would be glad to help you optimize your server on a commercial basis. If you would like free help, please try forum.percona.com instead of posting into the comments here.

By: Gabek

Gabek — Thu, 02 Dec 2010 08:24:37 +0000

Hi,

I am not an expert at mysql.

Can you help me optimize my server?
(Might give me a conf file example?)

Thies are te informations I have about mysql:

[Ed: deleted output of a tuning script]

By: Baron Schwartz

Baron Schwartz — Thu, 18 Mar 2010 17:43:35 +0000

Yes, it will be accurate and still meaningless

By: CÃ©dric

CÃ©dric — Thu, 18 Mar 2010 17:21:24 +0000

But if you retrieve a ratio every 15 minutes, you don’t need to know if the server has been restarted, and you have a more accurate ratio !

By: Baron Schwartz

Baron Schwartz — Thu, 11 Mar 2010 14:32:15 +0000

I don’t know. You changed two things at once: you restarted MySQL, and you changed its configuration. If you want to isolate the change to just one thing in the future you should use “SET GLOBAL key_buffer_size=512 * 1024 * 1024″ next time.

By: Andrew

Andrew — Wed, 10 Mar 2010 23:22:55 +0000

Thanks for the information. Interestingly, I decided to use suggestion 4: “Set key_buffer_size really big”.

Well, I upped the size from 256MB to 512MB and interestingly, when I look at Memory Used through WebHost Manager, I expected the memory used to increase. It had been showing at about 44% prior to changing the key_buffer, but immediately after doubling key_buffer and restarting mysql, Memory Used has been showing no higher than 25%.

Any idea why?

By: rguinot

rguinot — Thu, 04 Mar 2010 21:22:37 +0000

Thanks!
I guess we can then rely on system tools such as iostat or sar to tell if disk overall io is being saturated

By: Baron Schwartz

Baron Schwartz — Thu, 04 Mar 2010 21:12:58 +0000

Correction: the write IO is largely done in the background. The read IO is even more complicated in my view — there is read-ahead, for example, and a mixture of foreground and background IO.

By: Baron Schwartz

Baron Schwartz — Thu, 04 Mar 2010 17:57:42 +0000

You can find that in the docs. But InnoDB’s IO is largely done in the background and can rarely be traced directly to any user action with certainty.