> On 11/9/2010 4:31 PM, Simon Matter wrote: >>> On Mon, Nov 08, 2010 at 09:58:12PM -0500, Ashley Penney wrote: >>>> One thing to consider is the 'nobarrier' option for ext4 - apparently >>>> makes >>>> a big performance difference. >>> This is probably it. This was discussed ad infinitum when it was >>> changed upstream, and caused a huge performance regression across >>> pretty much every workload. >> But don't barriers affect mostly writes? If yes, is it because of the >> logs >> and atime? 80% still looks like too much, maybe it's so bad because of >> the >> virtualized environment? >> >> Simon >> > no because both were run in the same virtualized environment.
No, what I said is the virtualized environment possibly could make a much bigger difference with write barriers yes/no than it would on real hardware. As long as your virtualized server doesn't directly write to the disk it should be expected that write barrier settings behave much different than on real hardware. Simon _______________________________________________ rhelv6-beta-list mailing list rhelv6-beta-list@redhat.com https://www.redhat.com/mailman/listinfo/rhelv6-beta-list