Choosing proper innodb_log_file_size

If you’re doing significant amount of writes to Innodb tables decent size of innodb_log_file_size is important for MySQL Performance. However setting it too large will increase recovery time, so in case of MySQL crash or power failure it may take long time before MySQL Server is operational again.

So how to find the optimal combination ?

First let me explain what happens on recovery and why large innodb_log_file_size slows down recovery. During startup after crash Innodb scans log files to find log records which only have been applied in memory and do not exist in tablespace. Log records for modifications which did not make it to the tablespace are then applied. This is called redo phase of recovery. It can take pretty long time and this time depends on number of variables – how large are rows ? (smaller log records mean more records for same sized logs), how random were data modifications (random updates will need random IO to check if pages are up to date), number of unflushed pages in innodb buffer pool and its size as well as performance of IO subsystem. As there are so many factors, it is hard to come up with any general guidelines, something like 1GB per 10 minutes of recovery time – instead you would need to apply load which is typical for your application, crash MySQL in the middle and watch it to recover. Doing this several times you should be able to estimate how long recovery time take and adjust your logs apropriately. The good thing is – redo phase is close to be proportional to size of log files, so expect 1GB logs to take twice time to apply compared to 512MB logs.

Redo phase is however only one of the phases of recovery. The other important one is undo phase – after log file are applied and database is in “physically consistent” state, Innodb will need to roll back certain transactions which were not commited, but changes from which already made it to the database. Unlike “redo” phase “undo” phase can’t be reduced by sizing your log files. Even more undo phase can be slower with small log files. Undo phase takes considerable time if tranactions are long – Ie if you would delete 10000000 rows in the same transaction and crash in the middle recovery can take quite a long time. The only way you can reduce “undo” phase is size your transactions appropriately – so updates/inserts/deletes can be sized to affect limited number of rows.

Good thing about updo phase however is – it can be done in background as in MySQL 5.0. The rows affected by background rollback however might not be modified until rollback is complete.

One more thing to consider – how large log files do you need at all ? You could run benchmark with 1GB log files and 2GB and see if there is any performance benefit. After certain size increasing log file size do not dramatically increase performance, however this again depends on configuration and workload.

Note at this poing 4GB is maximum combined size allowed for innodb log files, which is however large enough limit for most onfigurations.

Q:I am a little confused by this line:
Note at this poing 4GB is maximum combined size allowed for innodb log files

My question is what is combined to get this 4GB maximum? Is it innodb_log_files_in_group * innodb_log_file_size or is there another part of equation I am missing?

A:It is combined size which is limited.

innodb_log_file_size

Command-Line Format	--innodb_log_file_size=#
System Variable	Name	innodb_log_file_size
	Variable Scope	Global
	Dynamic Variable	No
Permitted Values (<= 5.6.2)	Type	numeric
	Default	5242880
	Min Value	1048576
	Max Value	4GB / innodb_log_files_in_group
Permitted Values (>= 5.6.3, <= 5.6.7)	Type	numeric
	Default	5242880
	Min Value	1048576
	Max Value	512GB / innodb_log_files_in_group
Permitted Values (>= 5.6.8)	Type	numeric
	Default	50331648
	Min Value	1048576
	Max Value	512GB / innodb_log_files_in_group

The size in bytes of each log file in a log group. The combined size of log files (innodb_log_file_size *innodb_log_files_in_group) cannot exceed a maximum value that is slightly less than 512GB. A pair of 255 GB log files, for example, would allow you to approach the limit but not exceed it. The default value is 48MB. Sensible values range from 1MB to 1/N-th of the size of the buffer pool, where N is the number of log files in the group. The larger the value, the less checkpoint flush activity is needed in the buffer pool, saving disk I/O. Larger log files also make crash recovery slower, although improvements to recovery performance in MySQL 5.5 and higher make the log file size less of a consideration. For general I/O tuning advice, see Section 8.5.7, “Optimizing InnoDB Disk I/O”.

参考：

http://www.percona.com/blog/2006/07/03/choosing-proper-innodb_log_file_size/

http://dev.mysql.com/doc/refman/5.6/en/innodb-parameters.html#sysvar_innodb_log_file_size