XFS Best Practices

[Jeff Flowers]

On Thu, Aug 20, 2009 at 11:23 AM, Eric Sandeen<sandeen at sandeen.net> wrote:
> Jeff Flowers wrote:
>> I am going to use XFS on a Arch Linux box and I am looking for ways to
>> maximize XFS performance. According to an article I have read [1],
>> best XFS performance was reached with a file system formatted with a
>> 64MB log and mounted with 8 log buffers and atime disabled. But I am
>> curious, from the prespective of the XFS experts of this list, if this
>> is still good advice and if it is still relevant, as this article was
>> published in 2003.
>
> Based on the information you've provided about the performance issues
> you're seeing with your particular workload (i.e., nothing), the
> existing defaults are perfect for you.  :)

For me, it is not about dissatisfaction with XFS performance but
simply wanting to know if there are optimizations I am missing and
could be taking advantage of. Many forums have people talking about
options to improve or optimize Ext3 and Ext4 performance but XFS seems
to be dismissed (which I don't understand, as XFS is a very mature and
proven filesystem).


>> Also, I have seen a few people recommend turning off the internal
>> buffers of hard drives (via hdparm) when using a file system like XFS.
>> Good advice?
>
> When drive write caches lose power it may lead to inconsistencies in a
> journaling filesystem like xfs, which relies on data hitting the disk in
> a certain order, more or less.  By default xfs issues barriers to
> enforce this ordering; this has the effect of flushing the write cache
> to make it safe.  In some cases disabling barriers and also disabling
> write cache may be a good choice.
>
> If you "never" lose power (good ups?) then write caching is safe even
> w/o barriers.
>
> -Eric

Thanks for the information. Your explaination of write barries is one
of the better ones I have read.

-- 
Jeff