MPITHRASH
---------

MPITHRASH is an MPI code which produces an large unpredictable MPI
load on the network and MPI subsystem.  It is designed to test the
scalability of the high performance interconnect for a general work
load.

Building
--------

 * Compile the code for the host platform, eg.
       f90 -fast mpithrash.f90 -o mpithrash -lmpi

 * Resolve any problems with the build

Running the benchmark
---------------------

 * Run the benchmark on 16, 32, 64, 128 and N processors (where N
   processors is the total number of processors being offered), eg.
       mpirun -np 16 ./mpithrash
       mpirun -np 32 ./mpithrash
       mpirun -np 64 ./mpithrash
       mpirun -np 128 ./mpithrash
       mpirun -np N ./mpithrash

 * Ensure that the maximum number of processors run is at least twice
   the size of the largest SMP compute node (note that the vendor is
   at least required to run the code upto 128 processors)

 * All the output should be provided

 * The TEST9.{16,32,64,128, N} result is the MB/s per process reported
   by mpithrash

Resoures
--------

 * Approximately 1GB per process

 * Each benchmark runs in a few minutes (depending on network performance)