Discussion

The code listing and the table of comparisons along with corresponding graphs are attached with this report. The contour plots for the 1, 500 and 1000 iteration cases are also attached. The parallel code speedup for fixed problem size seems to be following Amdahl's Law while the speedup for the scaled problem size seems to follow the Gustafson's Law. From the calculation of R assuming Amdahl's law holds for the fixed problem size case, we note that the implemented Jacobi code is approximately 90-92 percent parallel. The mflops per processor for the scaled problem case is seen to be almost constant at approximately 4.5 mflops. The communication time takes most of the time in the execution of the code (approximately 50-60 percent).