CMSC 714- High Performance Computing

Fall 2011 - MPI Programming Assignment

Due Wednesday, September 28, 2011 @ 6:00PM

The purpose of this programming assignment is to gain experience in parallel programming on a cluster and MPI. For this assignment you are to write a parallel implementation of a program to simulate the game of Life.

The game of life simulates simple cellular automata. The game is played on a rectangular board containing cells. At the start, some of the cells are occupied, the rest are empty. The game consists of constructing successive generations of the board. The rules for constructing the next generation from the previous one are:

    1. death: cells with 0,1,4,5,6,7, or 8 neighbors die (0,1 of loneliness and 4-8 of over population)
    2. survival: cells with 2 or 3 neighbors survive to the next generation.
    3. birth: an unoccupied cell with 3 neighbors becomes occupied in the next generation.

For this project the game board has finite size. The x-axis starts at 0 and ends at X_limit-1 (supplied on the command line). Likewise, the y-axis start at 0 and ends at Y_limit-1 (supplied on the command line).

INPUT

Your program should read in a file containing the coordinates of the initial cells. Sample files are located here and here. You can also find many other sample patterns on the web (use your favorite search engine on "game of life" and/or "Conway").

Your program should take four command line arguments: the name of the data file, the number of generations to iterate, X_limit, and Y_limit.

To be more specific, the command line of your program should be:

life <input file name> <# of generations> <X_limit> <Y_limit>

The number of processes the program will run on is specified as part of the mpirun command with the -np switch.

OUTPUT

Your program should print out one line (containing the x coordinate, a space, and then the y coordinate) for each occupied cell at the end of the last iteration.  The output should go to standard output, and no additional output should be sent to standard output.

Sample output files are available:

     MPI/life.data.1.100.250.250.out is the output of the file MPI/life.data.1 runs for 100 generations on a 250x250 board

     MPI/life.data.2.100.250.250.out is the output of the file MPI/life.data.2 runs for 100 generations on a 250x250 board

HINTS

The goal is not to write the most efficient implementation of Life, but rather to learn parallel programming with MPI.

Figure out how you will decompose the problem for parallel execution. Remember that MPI (at least the OpenMPI implementation) does not have great communication performance and so you will want to make message passing infrequent. Also, you will need to be concerned about load balancing.  To learn about decomposing the problem in different ways, you must generate two parallel versions of the program, one that uses a 1D decomposition (rows or columns) and one that uses a 2D decomposition (both rows and columns).

Once you have decided how to decompose the problem, write the sequential version first.

WHAT TO TURN IN

You must submit the sequential and both parallel versions of your program (please use file names that make it obvious which files correspond to which version) and the times to run the parallel versions on the input file final.data (for 1, 2, 4, 8 and 16 processes), running on a 500x500 board for 500 iterations.

You also must submit a short report about the results (1-2 pages) that explains:

GRADING

The project will be graded as follows:     

Item Pct
Correctly runs on 1 processor 15 %
Correctly runs on 16 processors 40% (20% each version)
Performance on 1 processor 10%
Speedup of parallel versions 20% (10% each version)
Writeup 15%

RUNNING MPI with OpenMPI on the bug cluster

See the file bugs.html for more information on using the OpenMPI implementation of MPI on the cluster.

The number of processes/processors your program will run with is specified as part of the mpirun command with the -np switch.

ADDITIONAL RESOURCES

For additional MPI information, see http://www.mpi-forum.org (MPI API) and http://www.open-mpi.org (for OpenMPI)

For basic, but somewhat out of date, information about using the Maryland cluster PBS scheduler, MPI, etc., see http://www.umiacs.umd.edu/labs/LPDC/classguide.htm . For more details about using the bug/hive cluster, see https://wiki.umiacs.umd.edu/umiacs/index.php/ClusterGuide , which includes an example of how to use OpenMPI for your project.