A2: OpenMP Assignment

First review the general instructions for assignments.

1 - Explore the OpenMP reference materials on the course web page. Select three interesting OpenMP directives that we didn’t discuss in class. Describe each one in a paragraph that describes what it does, why it is interesting, and how you might use it.

2 - Explore the benchmarks presented in class. Select three benchmarks to evaluate. One must be fractal, one must be chosen from (matrix, heat) and the other from (nbody, nqueens). For each benchmark, do the following:

3 - Establish a baseline. Read the code to understand the fundamental operation. Adjust SIZE, ITER, DELTAT so that the benchmark runs in about 60s on a single core. Make any adjustments needed to avoid excessive output.

4 - Create an improved version of the benchmark that uses OpenMP for parallelization. Be creative and make use of appropriate directives and capabilities, going beyond what we discussed in class. Take care to understand which data should be private/shared within threads. Test your solution using 1 to 4 threads (but no more) on the CRC Front End machine. Double check that the parallelized version produces correct output. Compile with -fopenmp to enable OpenMP, for example (replace benchmark.c with your source file):

g++ -O1 -fopenmp benchmark.c -o benchmark -lm

5 - Evaluate the performance of the benchmark on 1-64 cores, but don’t run on the front end. Instead, submit each execution using HTCondor, so that the job runs on a node in the cluster. For example, create a submit script benchmark.8.submit something like this:

universe = vanilla

executable = ./benchmark
environment = OMP_NUM_THREADS=8

output = benchmark.8.out
error = benchmark.8.err
log = benchmark.8.log

request_cpus = 8
request_memory = 1024MB
request_disk = 100MB

should_transfer_files = yes
when_to_transfer_output = on_exit

queue 1

Then submit the job like this:

condor_submit benchmark.8.submit

Remember the job ID printed by condor_submit, such as 12345.0. To see whether it is still in the queue, replace NETID with your NetID:

condor_q NETID

6 - Evaluate the performance of the benchmark on 64 cores as SIZE starts small and increases by powers of two. Stop if your runtime exceeds 30 minutes on 64 cores. As above, run these jobs on the cluster, not the front end node. Increase request_memory and request_disk if needed for larger SIZE values. (Note: There will be queueing delays while waiting for 64-core machines. Don’t wait until the last day to submit these jobs.)

7 - Plot your results from step 5 (vary cores) and step 6 (vary size) and discuss the results, being sure to point out and explain any unexpected behaviors. Take some time to think through the most clear and insightful way to present the results.

(Don’t forget to repeat for three benchmarks!)

Turning In

Commit all of your code, scripts, data, and plots to your course repository in a directory called openmp. Include a README.md that ties everything together and addresses the points above. In all things, show insight, curiosity, and craftsmanship. Be sure to push everything to GitHub! Turn in your work by submitting the URL of your repository to the corresponding assignment page in Canvas.