In my previous posts I highlighted that one of improvements in Percona Server is support of innodb_log_file_size > 4G. This test was done using Percona Server 5.5.7, but the same performance expected for InnoDB-plugin and MySQL 5.5.
The valid question how long is recovery in this case, so let’s test it. I took the same tpcc-mysql 1000W workload with 52GB and 144GB innodb_buffer_pool_size with data located on Virident tachIOn card and killed mysqld after 30 mins of work.
The recovery time after start is:
for 52GB innodb_buffer_pool_size:
|
1 2 3 |
101220 21:54:31 InnoDB: Database was not shut down normally! .. 101220 22:02:40 InnoDB: Rollback of non-prepared transactions completed |
that is 7min 51sec
for 144GB innodb_buffer_pool_size:
|
1 2 3 4 |
101220 22:45:37 InnoDB: Database was not shut down normally! .. 101220 22:55:00 InnoDB: Rollback of non-prepared transactions completed |
that is 9min 23sec
and
for 144GB innodb_buffer_pool_size with data stored on RAID10:
|
1 2 3 |
101220 23:46:01 InnoDB: Database was not shut down normally! .. 101221 0:00:58 InnoDB: Rollback of non-prepared transactions completed |
that is 14min 57sec
I think this time is acceptable as trade-off for performance.
However it should be taken into account if you use HA solution like DRBD, as it basically means this is time for fail-over period, and your system will be down during this time.
