Modeling and characterizing parallel computing performance on heterogeneous networks of workstations

A heterogeneous network of workstations (NOW) introduces a new performance factor into distributed computing: a large variation of the computing power of the different workstations. This unique factor makes traditional performance models/metrics for homogeneous computing measurement and evaluation not suitable for heterogeneous computing. We present models which quantify the heterogeneity of networks and characterize the performance effects. The models consider effects of both the heterogeneity and time-sharing in a nondedicated environment. Speedup, efficiency and scalability are defined. These models are general enough to cover performance evaluation of both homogeneous and heterogeneous computations in dedicated and nondedicated NOW systems. To validate and support performance modeling results, we conducted a collection of experimental measurements for evaluating computing performance and scalability of a group of application programs on a heterogeneous NOW