List:General Discussion« Previous MessageNext Message »
From:Nico Sabbi Date:April 9 2008 12:51pm
Subject:Help needed to tune Innodb on ZFS (on Solaris)
View as plain text  
Hi,
I installed Mysql 5.0.45 on Solaris 10 01/08 configured to run
the datadir on a ZFS filesystem on a Raid1 pool.
Both myisam and innodb tables are on the same filesystem.
Innodb is configured to run with a buffer_pool_size=256M,
with doublewrite set to OFF and with file_per_table to Off.
The log files are 2 x 48 MB each.
The problem I have is that importing the same 7 MB sql dump
takes 9 seconds if engine=Myisam and 98 when engine is Innodb.

Following some advice found in various bloggers I 
created the zfs filesystem I created with a recordsize of 16K, 
and set flush_log_at_trx_commit to 0. In some way the 
benchmark have improved, but I still find Innodb much slower than
Myisam.

Does anyone have any experience on this particular configuration?
Is there any other trick to follow to improve Innodb's performance
on ZFS?

Three more things that I noticed:
- setting innodb_flush_method=O_DIRECT causes mysql to 
complain that the directio() is not implemented.

- dropping the db and reimporting it from the dump is slower
at every iteration. Is there any way to keep the next reimports
as fast as the first one without recreating from scratch the ibdata 
files?

- why using innodb_file_per_tables is so much slower than 
running innodb with a single table space?

Thanks in advance for any help,
	Nico
Thread
Help needed to tune Innodb on ZFS (on Solaris)Nico Sabbi9 Apr
  • Re: Help needed to tune Innodb on ZFS (on Solaris)Rob Wultsch9 Apr
    • Re: Help needed to tune Innodb on ZFS (on Solaris)Nico Sabbi9 Apr
Re: Help needed to tune Innodb on ZFS (on Solaris)Rodolphe JOUANNET10 Apr