Hazim Shafi has written a very relevant case study illustrating how significantly memory usage patterns can affect speedups of parallel applications.  Have you ever parallelized an application only to find marginal (or no) speedup?  Perhaps this entry will explain why. 

In this entry, I will describe an important parallel programming concept, data parallelism. I will then show data parallelism in action by presenting a simplified implementation of the Jacobi Relaxation algorithm. Data Parallelism In general, data parallelism occurs in algorithms where the same set of...