Support for MPE logging (MPI calls and user-defined states/events)

[Lisandro Dalcin]

I've just finished to implement support for MPE logging:
http://www.mcs.anl.gov/research/projects/perfvis/download/index.htm

MPE is a standard component of MPICH2, so MPICH2 users can benefit
from this immediately (well, after building mpi4py from SVN). Open MPI
users will have to build MPE and configure mpi4py (using mpi.cfg, for
example) pointing to the location of MPE headers and libs...

Using MPE was always possible by building mpi4py with MPE, but now the
logging can be enabled at runtime, it does need of a recompile. I
mean, you use build and use mpi4py as always, but at some point you
are curious about what's going on with the MPI communications... Then
you just add a couple of lines to the main bootstrap script, and at
the end of your run you get a log file for post-mortem visualization
of the communication flow with the "jumpshot" viewer. So no need at
all to rebuild mpi4py to get MPI logging :-).

MPE also let users define their own states and events, so you can
easily instrument your own code. This is also possible in mpi4py. Just
take a look at this full working example:
http://code.google.com/p/mpi4py/source/browse/trunk/demo/mpe-logging/cpilog.py

PS: The logging of MPI calls will definitely NOT work on Windows, but
user-defined states and events should work. Regarding OS X, I've do
not know if it will work: enabling MPI profiling at runtime relies on
dynamic loading with dlopen() and appropriate mode flags (basically,
RTLD_GLOBAL); In OS X, Python ext modules (.so) are not the same as
shared libraries (.dylib), though perhaps this difference is not
noticeable in recent OS X. I hope it works (after all, we are talking
about "The world's most advanced OS").

Regards,

--
Lisandro Dalcín
---------------
Centro Internacional de Métodos Computacionales en Ingeniería (CIMEC)
Instituto de Desarrollo Tecnológico para la Industria Química (INTEC)
Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET)
PTLC - Güemes 3450, (3000) Santa Fe, Argentina
Tel/Fax: +54-(0)342-451.1594

[Brian Granger]

Lisandro,

This looks quite nice! I really like your usage of the with statement
in the example.
I will get you the account on my OS X box soon (just got the firewall
opened up) so
you can do more testing on OS X.

Cheers,

Brian

--
Brian E. Granger, Ph.D.
Assistant Professor of Physics
Cal Poly State University, San Luis Obispo
bgra...@calpoly.edu
elli...@gmail.com