[Gluster-users] Increased brick 'activity' when moving 3.2.5->3.3.0
david at davidcoulson.net
Sat Jun 2 00:19:42 UTC 2012
For what it is worth, here is a graph showing bandwidth utilization
between two of my four Gluster nodes following 3.2.5->3.3.0 upgrade.
There were no other changes to the environment or work load, other than
the Gluster update.
I'm going to try to reproduce it on another environment this weekend to
get more solid numbers.
On 6/1/12 12:28 PM, David Coulson wrote:
> I upgraded my 3.2.5 environment to 3.3.0 this morning. I'm seeing an approx 4x increase in network activity since the upgrade. tcpdump indicates a volume which is pretty much 100% reads has a lot of tcp activity between the nodes. Since it is mounted NFS, I was expecting that for a 'nearly all read' workload it would utilize the OS cache (this is what I experienced when we moved from Fuse to NFS under 3.2.5).
> Is there a volume tweak or a change to NFS I need to make to utilize OS caching? We're running on RHEL6. Here is my mount output for this volume.
> localhost:/ldirectord on /etc/ldirectord type nfs (rw,noatime,nodiratime,tcp,vers=3,nolock,addr=127.0.0.1)
More information about the Gluster-users