Locking impacts and Loadbalancing
When using a gfs-cluster with multiple nodes being used as active/active cluster with some kind of loadbalancer in front or integrated these hints might help.
Locking with DLM
As later versions of GFS with DLM as lockmanager cache locks not aquired by other nodes it can be a huge performance boost to direct clients always to the same gfsnode. Then fewer locks possesed by other nodes have to be aquired. Each request are some small packets sent over the locking network.
That means it often has great performance impacts to loadbalance persistently. Which means every client is directed to the same node ever an ever again.
Example:
IMAP can benefit very much from such configuration cause normally every client accesses the same directory or files wheras all other access other files or directories. When all requests are balanced unpersistently loads of locks have to be transfered from one node to the other. When requests are balanced persitently every client comes back to the same node and less lock go over network.
ATTENTION:
This might not be the case for any application but is often worse thinking about.