[Gluster-users] number of bricks given doesn't match required count

[Kaleb S. KEITHLEY]

On 02/14/2014 10:44 AM, Florent Bautista wrote:
> Hi all,
>
> I would like to try GlusterFS but I really don't understand how is
> handled stripping and replication.
>
> I have 3 nodes, with 1 brick each.
>
> What can I do with that ?
>
> I would like files stripped across the 3 servers and also files
> replicated on the 3 servers. Why not can I do that ?
>
> Then, why when adding new nodes, do we have to set multiple of {I don't
> know which count} ? I mean, I want 3 replicas of every files, I have 4
> servers, what is the problem ? Why GlusterFS can't balance replicas
> across the available nodes ?
>
> I could understand that putting 4 replicas into 3 nodes is problematic,
> but 3 in 4 I don't understand.

With 4 bricks (brick = node + volume) and "replica 3" you'd only have a 
replica of one of your bricks, the other two would not be protected.

But let's imagine that your first three bricks are 1TB disks, and the 
fourth is a 3TB disk. You could split the 3TB disk, e.g. with LVM, into 
three volumes to use as bricks to give you the full ×3 replication 
required by gluster replication.

Also Gluster's stripe is not the same as RAID-0. While similar, it 
doesn't give the performance improvement that real RAID-0 striping 
gives. I'd use gluster's stripe only if I have files that are so large 
they won't fit on a single brick. If you just want greater capacity for 
regular files you should use distribute (dht).

-- 

Kaleb