Extreme slowness with xfs [WAS: Re: Slowness with new pc]

[Dave Chinner]

On Wed, Nov 26, 2008 at 01:30:26PM +0100, Martin Steigerwald wrote:
> I wonder whether XFS performance with barriers enabled can be improved? 

Not really. It requires fundamentally altering the way the
transaction system works, and that's no easy task. I have a few
things that I'm looking at, but nothing that would be considered
short-term....

> And whether XFS with disabled write cache (via hdparm) but without 
> barriers might even be *faster* than XFS with barriers... One thing to 
> test eventually.

Yes, it often is with SATA drives - it depends on the quality of
the NCQ implementation in the drive. For SCSI drives, no write
cache or barriers is almost always faster than using WC+barriers.

Cheers,

Dave.
-- 
Dave Chinner
david at fromorbit.com