A critical look at three of parallel computing´s maxims

The last few years have seen the introduction of a large number of parallel computers, as well as the failures of several manufacturers. As the machines have been released, they have been thoroughly evaluated, and a wide range of benchmark data is now available. Furthermore, in this time our expertise in programming and extracting performance from parallel machines has greatly increased, a theory of parallel and network algorithms has developed, and our understanding of the modeling process has improved. Hence, now seems a propitious time to take a critical look at our maxims. This paper takes a critical look at the following three maxims: parallel architecture is converging on a design based on commodity microprocessor chips; wormhole routing is decidedly more efficient than store-and-forward routing; and the PRAM is an unrealistically ideal model of computation