Progress in Parallel Algorithms

Abstract

Parallel computing offers the promise of increased performance over sequential computing, and parallel algorithms are one of its key components. There has been no aggregated or generalized comparative analysis of parallel algorithms. In this thesis, we investigate this field as a whole. We aim to understand the trends in algorithmic progress, improvement patterns, and the importance and interactions of various commonly used metrics. We collect parallel algorithms solving problems in our set and analyze them. We look at four major themes: how parallel algorithms have progressed, including in relationship to sequential algorithms and parallel hardware; how the work and span of algorithms influence performance; how problem size and available parallelism affect performance; and what researchers’ observable priorities look like. We find that more problems have had parallel improvements than sequential ones since the ’80s, that most parallel algorithms don’t improve algorithmic complexities, and much more. This research is important for us to understand how the field of parallel algorithms has changed throughout time, and what it looks like now.