Heikki Tuuri wrote:
>
> A journaling file system like ReiserFS does not help if fsync does not
> work. A journaling file system itself is actually a bit like a
> transactional database. A broken fsync might cause bad damage there.
>
> I would be happy if users tested the 'pull-the-plug' performance of
> Linux-2.6.10/InnoDB. Jens Axboe might have solved most fsync problems:
>
> http://groups-beta.google.com/group/linux.kernel/browse_frm/thread/bbe45994b0277f7a/cc6d86c50514da81?q=axboe+fsync+linux&_done=%2Fgroups%3Fas_q%3Daxboe+fsync+linux%26safe%3Dimages%26as_scoring%3Dd%26lr%3D%26hl%3Den%26&_doneTitle=Back+to+Search&&d#cc6d86c50514da81
>
I am running 2.6.10 here. I will see if I can set up a test case this
weekend.
Also, I know you and others have mentioned that Linux 2.6 + Opteron +
Innodb is a problematic situation. Could you expand on this? From our
personal experience (running mysql on Opteron + linux 2.6.10 w/ myisam
tables), we have seen very slow performance when running intensive IO
operations (deleting 20 million rows from a 50 col table) and we have
experienced a greater number of index corruption on the opterons than on
our intel dbs.
Greg
--
greg@stripped
Meetup.com
