Corey Kovacs wrote: > Also, you might consider using noop as an IO scheduler (or lack of one) > since large SAN's (EVA8x00) handle command reordering at the controller > level. This can make a difference and you can change it on the fly as > well.
That was my assumption, too. But when I measured it (with bonnie++ and fio), I found that the cfq scheduler actually gave a bit better performance. About queue depths: With four paths behind a file system (active-active-active-active round- robin), I found that going from the default queue depth (32) to 192, I got a significant performance increase. But going further didn't yield much, so I've stayed at 192. If John Haxby is right, this means that I allow up to 4x192=798 commands to be queued at the storage system. By the way: The storage system is XIV. Filesystem: ext3 with a journal size of 256. -- Regards, Troels Arvin <[email protected]> http://troels.arvin.dk/ _______________________________________________ rhelv5-list mailing list [email protected] https://www.redhat.com/mailman/listinfo/rhelv5-list
