Hi Jens,
If I recall correctly, your observation lies in the type of
hardware you are using. I think some hardware allows a developer to
leverage an interrupt while some requires polling for a received
message. Design requirements for MPI to be fast with low latency
usually (I presume) outweigh lowering heat production.
I think I first noticed this issue with mvapich when infiniband was
new. I don't think we had it with Myrinet 2000 or Gig-E.
I also think ParaView is a normal mpi application, we just don't
notice machines spinning while waiting during batch processing, we
just assume they're working hard.
-John.
On Dec 5, 2008, at 9:42 AM, Jens wrote:
Hi John,
thanks for your answer. That makes sense. "Normal" mpi-apps are
probably
not written to wait for more things to do - they are simply always
busy.
It is just a pity that the cluster has to run 100% producing a lot of
heat for nothing.
So the MPI-lib will probably not change this behavior :( ? (I am using
open-mpi 1.2.8)
Greetings
Jens
John M. Patchett schrieb:
Hi Jens,
Your pvserver is probably waiting on an MPI_Recv and your MPI
implementation is spinning.
You will note that process 0 probably isn't doing this, as the other
nodes are waiting on process 0 to send.
I have searched this problem all the way to the MPI developers as
it's
easy to replicate without paraview and the MPI guys assure me the
alternatives are worse.
-John.
On Dec 5, 2008, at 8:42 AM, Jens wrote:
Hi,
if I run "mpirun -np 4 ./pvserver" on our cluster-node and
connect from
my client, this pvserver always shows 100% cpu usage - even if I do
nothing at the client.
Seems to me as if there is a loop waiting for the client to ask for
action - but this loop is calling no wait/sleep function.
Greetings
Jens
_______________________________________________
ParaView mailing list
[email protected]
http://www.paraview.org/mailman/listinfo/paraview
_______________________________________________
ParaView mailing list
[email protected]
http://www.paraview.org/mailman/listinfo/paraview
