Computer scientists at Lawrence Berkeley National Laboratory and the University of California at Berkeley wrote this paper for the 2003 International Parallel and Distributed Processing Symposium, with the intention of evaluating five different systems and the performance of their respective networks. Advantages of each configuration are outlined, and some optimization techniques are suggested. Expanding on the concept of parallel operation, the authors propose overlapping high-latency communication with other necessary tasks to maximize efficiency. The evaluations were conducted using a set of benchmarks to measure accurately performance parameters.
Comments