[Users] Why choose glusterfs over iscsi for oVirt storage?

[Steve Dainard]

On Mon, Feb 17, 2014 at 3:50 AM, Justin Clacherty <justin at redfish.com.au>wrote:

> Hi,
>
>
>
> I'm just setting up some storage for use with oVirt and am wondering why I
> might choose glusterfs rather than just exporting a raid array as iscsi.
> The oVirt team seem to be pushing gluster (though that could just be
> because it's a Red Hat product).  Can anyone answer this one?
>

If you're contemplating software iscsi then you probably aren't working on
a production setup that requires high availability or data consistency. If
you are, you should take a deep dive on what your single point of failures
are. If you're actually talking about a SAN lun, then I would imagine you
aren't the storage admin, and can leave the particulars in their hands.

Its funny, in RH's current supported version of RHEV, native gluster
storage isn't even an option. This is a pretty new project, and if you
follow the bread crumbs on distributed file systems like glusterfs or ceph
you'll start to see the future advantages. You can also figure that seeing
as both projects are spearheaded by RH dev's that there is likely a lot of
cross-talk and feature requests making both projects better, and who
wouldn't promote a better solution?


>
>
> What I have come up with is as follows.
>
>
>
> For:
>

There is too much to cover in any of these advantages, you're going to need
to do a lot of research on each of these 'features' if you want to use them
successfully.

> -          easy to expand
>
> -          redundancy across storage devices/easy replication
>
> -          high availablility
>
> -          performance
>
> -          it's kind of cool J
>
> -          maintenance?
>
>
>
> Against (these are guesses):
>
> -          performance? (multiple layers of filesystems everywhere - fs
> in vm + image on gluster + gluster + block filesystem)
>
Its worse than just a ton of software layers. NAS protocols seem to be
focused on data consistency, which means they don't write async to storage.
iscsi is typically async and has much better throughput, but also a greater
chance for data loss or corruption. Technically you can achieve the same
level of performance as iscsi using NFS (backed by glusterfs if you like)
but you need to set options on the volume to allow async writes.


> -          complexity
>
If you're doing storage properly, the underpinnings are always complex
unless you're paying someone else to do it (read: SAN / managed service
provider). Research multipath and HA on software iscsi and you'll see what
I mean.


> -          maintenance?
>
>
>
> Any help here is appreciated.  Also, does the underlying block level
> filesystem matter here?  VMs running under ovirt would be typical business
> applications - file serving (samba4), email, databases, web servers, etc.
>

There is a lot to answer here and I don't have all the answers. Take a look
at the gluster docs for underlying file system requirements. Any block
device should work. Specifically I'll mention that the glusterfs team
doesn't suggest hosting db's on glusterfs - many small reads/writes are not
one of glusters strong points.


>
>
> Cheers,
>
> Justin.
>
>
>
>
>
>
>
> _______________________________________________
> Users mailing list
> Users at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20140220/318824cf/attachment-0001.html>