Re: MySQL optimization for testing purpose

[Frederic Descamps]

Hi Sebastien,

Several info are missing here, like hardware, configuration, etc...

However, you can always relax on durability for InnoDB and also avoid flushing to the logs to often (basic innodb stuff), you can also disable binary logs for example.

The variables you mentioned previously won't affect this to be honest.

so if you really don't care about your data --WARNING NOT SAFE FOR PRODUCTION-- you could try with these variables:

innodb_doublewrite=0
innodb_log_file_size=(double it)
innodb_flush_log_at_trx_commit=0
innodb_buffer_pool_size(increase it if you can)


if you have fast storage (so you don't rely on fs):
innodb_flush_method=O_DIRECT

Additionally, if during your testing you are doing it in single thread, and you don't bind to numa node 0 (i.e. socket 0) you could get different runs time depending on if there is cross node communication to complete the request. cross node coms can cause/cost as much as a 40% difference in time results.

These are just some ideas because as said earlier, much more info is required to provide you a decent answer ;)

lefred/