Gluster users - Jun 2012 - Increased brick 'activity' when moving 3.2.5->3.3.0

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)

Thanks-
David

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.

http://i.imgur.com/VnZPl.png

I'm going to try to reproduce it on another environment this weekend to 
get more solid numbers.

David

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)
>
> Thanks-
> David