On Mon, Jun 18, 2012 at 09:01:02AM +0000, Fernando Frediani (Qube) wrote: > Given that Gluster is not for high performance (low > latency) applications, RAID 5 seems to be a good option Remember that low latency and high throughput are two different things. RAID5 will give you low throughput as well as high latency, unless you happen to be writing large stripes at a time. > Hopefully for reads Gluster should be able to read > round-robin from both copies to improve throughput. I believe so, and if the client happens to be one of the nodes it also prefers the local node. > Unfortunately Gluster > doesn't yet have the ability to be aware of bricks on the same node and > don't put data there. As I've said twice before now, you *can* do this just fine. You simply configure the bricks in the correct sequence. gluster volume create foo replica 2 \ server1:/brick1 server2:/brick1 \ server1:/brick2 server2:/brick2 \ server1:/brick3 server2:/brick3 This is clearly documented at http://download.gluster.com/pub/gluster/glusterfs/3.2/Documentation/AG/html/sect-Administration_Guide--Setting_Volumes-Distributed_Replicated.html So dividing one server node into multiple RAID arrays is just fine.
