[Gluster-users] Why is xfs recommended?

[Lindsay Mathieson]

On Wed, 12 Nov 2014 10:24:34 AM Ravishankar N wrote:
> XFS scales well when there is lot of meta data and multi-threaded I/O
> involved [1]. Choosing a file system is mostly about running the  kind of
> workload you would expect your system to see, with your hardware
> configuration and your version of the OS.  If ext4 gives you better
> performance when used as back end for gluster with your settings and
> workload, there shouldn't be any reason why you cannot go with it.

In the end I went with Option B :) - ZFS. Built in support for journal and 
data ssd caches, and of course, all the ZFS goodies, which will be very useful 
in the future.

I tried dmCache, bcache and EnhanceIO with xfs and ext4. bcache was a buggy 
mess, dmcache was a pain in the ass to manage, EnhanceIO was pretty good. But 
I managed to generate data corruptions with all three.

Also disk benchmarks (bonnie++, dd, CrystalDiskMark) gave wildly varying 
results that didn't bear much relationship to observed improvements. 

In the end I went with a series of app benchmarks that matched our usage. For 
those all of the above including zfs gave good improvements in roughly the 
same range.

So all things being equal, zfs won out for being a well supported standard and 
all its other benefits.

thanks for all the help and advise folks, sorry for being a pain :) though no 
doubt I will continue to be so.

Cheers, from sunny BrizVegas.

-- 
Lindsay
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part.
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20141115/605c187f/attachment.sig>