I would like to understand in more detail how much time some collective
communication calls really spend waiting for the last process to enter. I
know this can be done by logging entry times for each process, but I
wonder if there is a better and more efficient way.
The peruse interface allows exactly that for send/recv communication (see
the current peruse specification
(http://www.mpi-peruse.org/current_peruse_spec.pdf) does not cover
collectives (but mentions them explicitly for future extensions).
Does anyone know if there is a plan to go ahead with MPI peruse for
collectives (in general and specifically in OpenMPI)?
Are there hooks in the OpenMPI collective's code which could be used to
hand-wire similar functionality?
Manfred MÃ¼cke rlcta.univie.ac.at
Research Lab Computational Technologies and Applications