Gluster users - Mar 2015 - GlusterFS vs NFS mount question

I'm looking at setting up a new Gluster Cluster, and I have a question
regarding mounting the volumes via clients...

According to what I've read, the documentation varies, or doesn't cover
this in the best detail.

It says the native glusterFS will provide the best concurrency.  And I've
read if you are accessing lots of small files NFS is the best option.

My question is, if I choose to use NFS to mount the volumes on client machines,
will this support concurrency amongst all the nodes, or when you provide the IP
address of the storage node as part of the mount option in /etc/fstab or mount
command, will all reads and write operations from that client hit just the
server listed in the mountpoint?

Basically, if I use NFS, do I then need to remember to have the clients point to
the different IPs of the physical storage nodes to balance IO, or do I need to
setup a proxy of some sort for load balancing, or is gluster taking care of
concurrency auto-magically even if I use NFS?

Thank you,
Taylor


-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://www.gluster.org/pipermail/gluster-users/attachments/20150318/e41acd93/attachment.html>

On Wed, Mar 18, 2015 at 01:58:42PM +0000, Taylor Lewick
wrote:
> I'm looking at setting up a new Gluster Cluster, and I have a question
> regarding mounting the volumes via clients...
> 
> According to what I've read, the documentation varies, or doesn't
> cover this in the best detail.
> 
> It says the native glusterFS will provide the best concurrency.  And
> I've read if you are accessing lots of small files NFS is the best
> option.
> 
> My question is, if I choose to use NFS to mount the volumes on client
> machines, will this support concurrency amongst all the nodes, or when
> you provide the IP address of the storage node as part of the mount
> option in /etc/fstab or mount command, will all reads and write
> operations from that client hit just the server listed in the
> mountpoint?
> 
> Basically, if I use NFS, do I then need to remember to have the
> clients point to the different IPs of the physical storage nodes to
> balance IO, or do I need to setup a proxy of some sort for load
> balancing, or is gluster taking care of concurrency auto-magically
> even if I use NFS?
When an NFS-client mounts a volume from an NFS-server, all the I/O will
happen through that particular NFS-server. The NFSv3 protocol does not
have a feature where multiple NFS-server can be used simultaneously.

The FUSE client will contact the storage servers that (should) have the
file it wants to use. Because different files are located on different
bricks (and often servers), this automatically distributes the load.

In the future, NFSv4.1 with pNFS support will become available through
the integration with NFS-Ganesha. pNFS makes it possible to contact
multiple NFS-servers (DS in pNFS speak). This should become usable after
glusterfs-3.7 has been released. Until then, you are stuck with NFSv3 or
the FUSE client.

If you have many NFS-clients, you can distribute the load a little by
using Round-Robin DNS or similar.

HTH,
Niels
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 181 bytes
Desc: not available
URL:
<http://www.gluster.org/pipermail/gluster-users/attachments/20150318/66c84426/attachment.sig>