[PVE-User] GlusterFS Questions

[Angel Docampo]

On 23/10/14 03:32, Lindsay Mathieson wrote:
> I'm playing around with GlusterFS and have a few questions. Have got a 
> test setup running on a 2 node cluster (using external USB Drives!) 
> that works surprisingly well :) and I must say it was very easy to setup.
>
>
> 1. I wanted to implement a distributed/replicated file system. Is 
> GlusterFS the best choice for a 2 node system? Would it be easy to add 
> a third node later?
>
>
> 2. When launching a VM off Node 1 or Node 2, is it accessing the local 
> hard disk directly, or does it all go through a network share? Our 
> network is only 1GB and it would be nice to take some of the load off it.
>
>
> Should I be asking these question on the forum rather than the list?
>
>
> Thanks,
> -- 
> Lindsay
>
>
> _______________________________________________
> pve-user mailing list
> pve-user at pve.proxmox.com
> http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
I'm using successfully Proxmox with GlusterFS. Each proxmox node is also 
a gluster node, and now I have two nodes (using a third proxmox node on 
vmware for proxmox quorum). I must say it works pretty nice, I even 
provide Windows VDIs with this system. Adding a third (with a fourth 
-it's a replica, you know-) and more nodes, it's pretty straight 
forward. Just prepare them, add to the gluster and the data will be 
rebalanced, with a 10Gb network, the data will be rebalanced hugely 
faster than with a 1Gb network.

I also suggest to you to use a better network because glusterfs writes 
are synchronous. The benefit of this is that if one node comes down, all 
the information is in the other node, so there are chances that the VMs 
even didn't notice when a node comes down.

When you write on gluster, even you have mounted it using the IP of one 
of them, the writes are "randomly" (not really random) send to any node, 
and then, replicated amongst them, it's the internal operation.

You could set up a georeplica instead a replica, designed to replicate 
two gluster clusters amongst remote offices, and then you could use your 
1Gb network to make asynchronous writes. The cons I see are you don't 
have *exactly* the same data on the both clusters at each time, and in 
case of a node comes down, you should reconfigure all VMs to point their 
disks to the other GlusterFS, because they are two different clusters.

So, summarizing: Yes, you can use gluster in a two-node gluster setup, I 
do not know if Ceph its better though, I can only share my positive 
experiences with gluster. I would recommend you to use a 10Gb network 
dedicated for gluster communication, it will be useful to replicate data 
synchronously and essential for rebalance the data when in a future you 
will need to add more nodes.

-- 

	
*Angel Docampo
*
*Datalab Tecnologia, s.a.*
Castillejos, 352 - 08025 Barcelona
Tel. 93 476 69 14 - Ext: 706
Mob. 670.299.381

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.proxmox.com/pipermail/pve-user/attachments/20141023/664d6eb9/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Logo-DLtec.png
Type: image/png
Size: 5898 bytes
Desc: not available
URL: <http://lists.proxmox.com/pipermail/pve-user/attachments/20141023/664d6eb9/attachment.png>