[mpich-discuss] mpich2 hangs on Ubuntu beowulf cluster(with NFS)

Nicolas Rosner nrosner at gmail.com
Wed Jan 4 17:26:42 CST 2012


Agree with Gus: looks like faulty code on an otherwise working setup.

Besides posting some code, if you're stuck and need help, you may want to try

- adding some print statements before/after relevant sends/recvs to
try and deduce where the program might be deadlocking

- a few web searches for e.g. `mpi deadlock {examples, causes,
avoiding}' or so, which should yield concrete examples and suggestions
to illustrate why/how/etc.

- reading up a bit about the various send modes, which (if you haven't
done so before) is important to realize some important causes of
frequent mistakes

 (e.g. not realizing that standard Send() may or may not block, or
that "blocking" may not mean what one'd think by extrapolation from
other, local-ony, contexts -- if you don't see what I mean here, then
it might indeed be useful to look up the MPI standard definition of
"blocking" and comparing with that of "synchronic").

HTH! N.


More information about the mpich-discuss mailing list