Compiling and running on the Hoffman Cluster

Compile:

In order to compile the code do as follows:

mpif77 program.f -o program

                                            (for Fortran code)

mpicc  program.c -o program

                                            (for C code)

Run:

Test run interactively : mpirun -np n executable

                                               ( will run your executable with n tasks)

Batch mode using mpi.queue after testing

                                              (typing mpi.queue on the command line will give you the self explanatory menu).

Excercises:

As a last resort you may look the Answer files.

1. Copy few of the following codes to your work directory and compile/run them on 2, 4 processors using mpirun as well as mpi.queue command.

pstruct: Fortran C

ptopb: Fortran C

ptopnb: Fortran C

barrier: Fortran C

bcast: Fortran C

scatter: Fortran C

gather: Fortran C

reduction: Fortran C

allreduction: Fortran C

derived: Fortran C

pack: Fortran C

group: Fortran C

poisson: Fortran C

 

2. Take the allreduction( Fortran C) code and do the reduction operation using  point to point calls

    (Answer: allreductionptop- Fortran C )

3. Take the following serial algorithm and implement it in parallel.

    (You may copy the serial version from dotproduct- Fortran C)

        Dot product of two arrays B(1000000) in to C(1000000)

                                                 DOT=SUM(B(i)*C(i)), i=1,1000000

       (Answer: pardotproduct - Fortran C)

4. Copy poisson (Fortran C) to your work directory and do the following:

Use MPI_SENDRECV function in place of send/recv separate calls  (Answer: poissonsendrecv-Fortran C )

Convert the non blocking send/recv calls in poisson: Fortran C to blocking send recv calls, and write down your experience.

Correct the code poisson- Fortran C with blocking calls in order it to work   (Answer: poissonblock- Fortran C )

Compare the timings for all these versions of runs for poisson and suggest the best one (timing comes in the output of batch job)