[Lustre-discuss] GlusterFS compared to Lustre

[Craig Tierney]

Nick Jennings wrote:
> Hey Guys,
> 
>   Sorry if this is the wrong place to post this, but I was wondering if 
> anyone has any thoughts on GlusterFS? It seems kind of like a user-space 
> version of Lustre, not cut out for huge clusters, but well suited for 
> smaller clusters without some of the resource requirements of Lustre.
> 
> I know it's not specifically Lustre related, but I don't know of a 
> better list of experts to post my questions to :)
> 
> -Nick
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss
> 

At our site, we have a 300TB Lustre filesystem pushing over 5GB/s served to over 600
clients.   I have tried GlusterFS and have a small filesystem using it now that
is accessible from a small subset of nodes, it doesn't get much load but
it was designed for that.  It is not intended to be a primary filesystem
to our clusters.

To me, Lustre is an optimized scalable parallel filesystem that is
works well at many scales, including the very large.  GlusterFS allows
you to build a distributed NAS, but the translator capability provides
much power and flexibility to filesystems that I haven't seen elsewhere.

Since GlusterFS is user-space (I think the client will one day be a kernel
module) and it runs of existing filesystems, it is much easier to get
running.  I have never tried to exercise it like our Lustre installation
to compare stability or scalability.

Craig
-- 
Craig Tierney (craig.tierney at noaa.gov)