[Gluster-users] 3.1 and performance over TCP transport

Peter Riel peterriel at me.com
Thu Nov 4 03:08:16 UTC 2010


Hi, this post was in regards to this statement, but as I don't want to pollute the original thread I'll start a new...

4 nov 2010 kl. 04.58 skrev Craig Carl:
> Samuele - 
> That happens automatically now. We do this so you can dynamically add and remove storage to your Gluster cluster, plus users no longer have to manage or edit config files. 


So with the new 3.1 volume file system, is the option of tweaking performance settings not there anymore? I've spent a few days playing with the .vol files, but can't seem to get my write speeds up to more than 10mb/sec on a 2 brick replicate cluster, this is using 3.1 on centos and measured from a workstation with a single nic 1Gbit. 
This troubles me as both bricks are running dual 1gbit nic's (bonded, rr) and internally they are connected with infiniband (all over TCP transport). As for read speeds from once of the workstations I get like max 80mb/sec (again tested from a workstation with a single 1Gbit nic).

Peter.





More information about the Gluster-users mailing list