Performance Comparison Initially, I created a fournode test platform with RPis. So, it was possible to do some preliminary testing of performance before creating the full cluster. The first test was to determine the general performance of a single node running multiple threads, using the MPI libraries. The MPI test program I used is one I developed during a parallel computing class at BSU. The program calculates pi using the Monte Carlo method. In this method, essentially the more random numbers you generate, the more accurate your resulting estimation of pi. It is an embarrassingly parallel way to calculate pi (i.e., scales easily to multiple independent processes, near 100% performance gain for each additional process added). All that is needed is a way to parallelize the generation of random numbers so that each process does not repeat the random sequence (remember, we are dealing with pseudorandom numbers here). In this case, I use a library called prand, developed by Dr. Jain and Jason Main at BSU.

