HI,
I can read on the manual in Section 9.1 (Errors-Common problems):
"In parallel, one way LAMMPS can hang is due to how different MPI implementations handle buffering of
messages. If the code hangs without an error message, it may be that you need to specify an MPI setting or two
(usually via an environment variable) to enable buffering or boost the sizes of messages that can be buffered".
Since in my simulation LAMMPS hangs pretty much as described , could you suggest me which variable to set in MPICH2?
Thanks
Fernando
HI,
I can read on the manual in Section 9.1 (Errors-Common problems):
"In parallel, one way LAMMPS can hang is due to how different MPI
implementations handle buffering of
messages. If the code hangs without an error message, it may be that you
need to specify an MPI setting or two
(usually via an environment variable) to enable buffering or boost the
sizes of messages that can be buffered".
Since in my simulation LAMMPS hangs pretty much as described , could you
suggest me which variable to set in MPICH2?
this is a question for the MPICH folks and not a LAMMPS question.
however, can you reproduce the issue with any of the
examples shipped with LAMMPS?
or can you try posting a small(!) input deck that reproduces
the problem that can be run elsewhere for testing/debugging?
most likely you are not running into an MPI issue, but simply
an issue of bad dynamics and one of your MPI processes going
bananas over it and getting lost. on x86 numerical instabilities
will not lead to coredumps and crashes, but the CPU will continue.
however, that often traps the code in some branch from which
it cannot return, since the code to test NaN or Inf doesn't exist.
axel.