Benjamin Slack benjamin.a.slack@wmich.edu CS5260 Assignment #2 Monte Carlo Estimation of PI using MPI
Compilation:
An accompanying Makefile will allow for complication of main.c into the a2 executable. I used the most recent MPI module on Thor for the complication environment. To compile the program, simple enter:
make
at the command prompt.
Running the batches:
I setup the a2.pbs file to run a single job. It expects a single parameter that must be specified in the qsub statement, the number of tosses to be executed. The node configuration is not supplied in the a2.pbs torque script. This configuration is supplied by the a2_batch.sh script, as a qsub parameter. Please be mindful of this fact and check the syntax that script uses before trying to execute a2.pbs directly.
The batches of jobs themselves are queued by running the a2_batch.sh shell script. This script will queue up batches of jobs for varying core configurations and numbers of tosses. I had some issues with overloading the Thor job submission queue, so the script will throttle itself and wait for groups of jobs to complete before starting the next batch. The longest batches are of course those only using a single processor, with wait times of 480 seconds. This value was determined experimentally in early testing. The maximum value is then halved to a minimum of 60 seconds between batches of more expansive resources requests.
Finally, once all requests in a given batch are complete, the shell script aggregates the output of the batch into a single CSV file for that batch. It then cleans up the working directory, eliminating all the outfiles and begins the next batch. The number of batches is controlled via a command line parameter of the a2_batch script of the following form:
a2_batch.sh
Each batch takes approximately 15 minutes, with all the waits, etc. So be mindful before queuing up a large number of them.
CMake:
The cmake files are for the use of my IDE (Jetbrains CLion). They are not required in the compliation outside of that IDE. The Makefile supplied should allow complication, so long as the OpenMPI module is loaded in the Thor session.