intel c++ compiler with mpich -- profiling

[jneeraj]

I am using Intel C++ compiler with MPICH library to compile my parallel code. I use gprof to profile the code.
As I incresae the number of processors for the parallel code, by profiling the code, I realize that the function
"sock_msg_avail_on_fd" consumes more and more time. Does anyone know what this function does and how can I reduce the time
spent on this function call.
Thank you,
neeraj

[TimP]

In principle, this question might fit better in the Clusters and HPC forum. Not knowing which version of MPICH you chose, I'd guess this is part of the implementation of an MPI_wait. You might see this, if you would look at mpicc pre-processed code. So, it may be a standard MPI performance tuning issue.

The subject of performance tuning MPI has been touched on in the Intel webinars on Intel Trace Analyzer and Collector. That tool was designed for problems such as you raise. ITAC should work with MPICH, at least if you chose a recent MPICH where thereis a compatible Intel MPI. The easy answer to your question, after checking that your communication drivers have optimum settings,is try Intel MPI.