CSCE Parallel Programming on a Multicore Multiprocessor Solution

Part 1. Shared-Memory Programming with Threads

Compile and execute the program in the file compute_pi.c, which computes an estimate of using the parallel algorithm discussed in class. It should be compiled and executed on grace.hprc.tamu.edu.

Load the Intel software stack prior to compiling and executing the code.

module load intel/2020a

To compile, use the command:

icc -o compute_pi.exe compute_pi.c -lpthread

To execute the program, use

./compute_pi.exe <n> <p>

where <n> represents the number of points and <p> represents the number of threads. The output of a sample run is shown below.

./compute_pi.exe 1000000 4

Trials = 1000000, Threads = 4, pi =    3.1433480000, error = 5.59e-04,

time (sec) =    0.0043

The run time of the code should be measured when it is executed in dedicated mode. Use the batch file compute_pi.grace_job to execute the code in dedicated mode using the following command on Grace:

sbatch compute_pi.grace_job



    1. Execute the code for n=108 with p chosen to be 2k, for k = 0, 1, …, 13. Using the experimental data obtained from these experiments, answer the following questions. For plots, use a logarithmic scale for the x-axis.

1.1. (10 points) Plot execution time versus p to demonstrate how time varies with the number of threads.

1.2. (10 points) Plot speedup versus p to demonstrate the change in speedup with p.
1.3. (5 points) Using the definition: efficiency = speedup/p, plot efficiency versus p to

demonstrate how efficiency changes as the number of threads are increased. 1.4. (5 points) In your experiments, what value of p minimizes the parallel runtime?

    2. Repeat the experiments with n=1010 to obtain the execution time for p=2k, for k = 0, 1, …, 13. 2.1. (5 points) In this case, what value of p minimizes the parallel runtime?

2.2. (5 points) Do you expect the runtime to increase as p is increased beyond a certain value? If so, why? And is this observed in your experiments.

    3. (5 points) Do you expect that there would be a difference in the number of threads needed to obtain the minimum execution time for two values of n? Is this observed in your experiments.




Page 1 of 3
CSCE 735 Fall 2021


    4. (5 points) Plot error versus n to illustrate accuracy of the algorithm as a function of n. You may have to run experiments with different values of n; for example n could be chosen to be 10k, for k = 3, …, 9. Use p = 48.


Part 2. Distributed-Memory Programming with MPI

Compile and execute the program in the file compute_pi_mpi.c, which computes an estimate of using the parallel algorithm discussed in class. It should be compiled and executed grace.hprc.tamu.edu.

Load the Intel software stack prior to compiling and executing the code.

module load intel/2020b

To compile, use the command:

mpiicc -o compute_pi_mpi.exe compute_pi_mpi.c

To execute the program, use

mpirun –np <p> ./compute_pi_mpi.exe <n>

where <n> represents the number of intervals and <p> represents the number of processes. The output of a sample run is shown below.

mpirun -np 4 compute_pi_mpi.exe 100000000

n = 100000000, p = 4, pi = 3.1415926535897749, relative error = 5.80e-

15, time (sec) =    0.0608

The run time of the code should be measured when it is executed in dedicated mode. Use the batch file compute_pi_mpi.grace_job, to execute the code in dedicated mode using the following command on Grace:

sbatch compute_pi_mpi.grace_job

    5. Execute the code for n=108 with p chosen to be 2k, for k = 0, 1, …, 6. Specify ntasks-per-node=4 in the job file. Using the experimental data obtained from these experiments, answer the following questions. For plots, use a logarithmic scale for the x-axis.

5.1. (10 points) Plot execution time versus p to demonstrate how time varies with the number of processes.

5.2. (10 points) Plot speedup versus p to demonstrate the change in speedup with p.

5.3. (5 points) Using the definition: efficiency = speedup/p, plot efficiency versus p to demonstrate how efficiency changes as the number of processes is increased.

5.4. (5 points) What value of p minimizes the parallel runtime?

    6. (10 points) With n=1010 and p=64, determine the value of ntasks-per-node that minimizes the total_time. Plot time versus ntasks-per-node to illustrate your experimental results for this question.

    7. Execute the code with p=64 for n=102, 104, 106 and 108, with ntasks-per-node=4.

7.1. (5 points) Plot the speedup observed as a function of n on p=64 w.r.t. p=1. You will need to obtain execution time on p=1 for n=102, 104, 106 and 108.


Page 2 of 3
CSCE 735 Fall 2021


7.2. (5 points) Plot the relative error versus n to illustrate the accuracy of the algorithm as a function of n.

Submission: Upload a single PDF or MSWord document with your answers to Canvas.





























































Page 3 of 3