[Beowulf] Performance tuning for Jumbo Frames

[Rahul Nabar]

I have seen a considerable performance boost for my codes by using
Jumbo Frames. But are there any systematic tools or strategies to
select the optimum MTU size? I have it set as 9000. (Of course, all
switiching hardware supports jumbo frames and no talking to the
external world required of the interfaces) Have you guys found
performance to be MTU sensitive?

Also, are there any switch side parameters that can affect the
performance of HPC codes? Specifically I was trying to run VASP which
is known to be latency sensitive. I have a 10 Gig E network with a
RDMA offload card and am getting average latencies (ping pong) using
rping of around 14 microsecs in the MPI tests. Is there a way to
figure out what percentage of this latency is in the switch and what
%age in the stack, cards and cables? Just trying to figure out which
are the battles one picks to fight.

Any tips?

-- 
Rahu