[Linux-cluster] performance using low-latency network for cluster/GFS communication

[Shawn Hood]

I was just curious if anyone has used IP-over-Infiniband or other
low-latency network configurations with GFS?  Can any conclusion be
drawn about how well this improves overall GFS performance, especially
in scenarios where the filesystems contain large numbers of small
files?

Along the same vein, how unadisable is it to use a shared gigabit
network that is mildly to moderarely utilized for these
communications?  It seems out network performance may be hurting our
overall cluster performance, causing nodes to drop out of the cluster.

Shawn Hood