Experiments with a Multiprocessor

Abstract

In this report we summarize four years of experience with the Multi-Maren multiprocessor laboratory.

We have preferred to work with problems concerning software, in particular whether there are algorithms that can utilize the potentially very large processing power of a multiprocessor.

We have found that this is certainly the case and in the report we present a few of these fast multiprocessor algorithms.

On the other hand we have also found a number of examples where the gain in using a multiprocessor is very small. In all these examples the limiting factors are inherent in the algorithm, they are not hardware bottlenecks. We justify this somewhat surprising claim by showing that even in multiprocessors with an order of magnitude more processors than ours it is unlikely that hardware bottlenecks will limit the performance.