[Pw_forum] Intensive load of cpu, mpich during pw.x or not?

[Sergey Lisenkov]

Dear PWscf authors and users,

I would like to ask from authors and other developers a little more information about running PWscf code in parallel. I use our new IBM cluster with MPICH-1.2.6 (GM) and Myrinet with gm driver (2.0.15). During different runs of the pw.x and after that when the code is finished, gm mapper on different nodes was killed. Our system people said that the problem is pw.x. I don't believe in that, because an average load of memory on each node is around 15%, no more. But they told me if the code writes a set of files (each processor writes), so thereis a intensive load network, very intensively uses a local disks (or GPFS server). Because in general I run 4-5 pw.x (in total 200 processors) there are some problems on the cluster. And I want to know is it really pw.x a source of the problems?

I wondering how frequently pw.x writes/reads data to/from the disks, how intensively it uses mpi?

All hints and suggestions are welcome.

 Best wishes,
   Sergey