I don't see you measure MPI_Allreduce. Basically you only measured some random numbers across processes. See osu_allreduce in OSU microbenchmark ( http://mvapich.cse.ohio-state.edu/benchmarks/) for a MPI_Allreduce benchmark. --Junchao Zhang On Mon, May 4, 2015 at 2:35 AM, David Froger <[email protected]> wrote:

Hello,

I've written a little MPI_Allreduce benchmarks [0].

The results with MPICH 1.4.1 [1] and Open MPI 1.8.4 [2] (1 column for the number of processors, and 1 columns for the corresponding wall clock time) show that clock time is half with twice processors.

But with `MPICH 3.1.4`, the wall clock time increase for 7, 8 or more processes [3].

In my real code and for all of the 3 above MPI implementation, I observe the same problem for 7, 8 or more processes, while I expect my code to be scallable to at least 8 or 16 processes.

So I'm trying to understand what could happen with the little benchmark and MPICH 3.1.4?

Thanks for reading.

Best regards, David

[0] https://github.com/dfroger/issue/blob/8b8bdd8e4b2b5e265c25fc2ba7077f6a108bb3... [1] https://github.com/dfroger/issue/blob/8b8bdd8e4b2b5e265c25fc2ba7077f6a108bb3... [2] https://github.com/dfroger/issue/blob/8b8bdd8e4b2b5e265c25fc2ba7077f6a108bb3... [3] https://github.com/dfroger/issue/blob/8b8bdd8e4b2b5e265c25fc2ba7077f6a108bb3... _______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

_______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss