Abstract:

The research of metrics for evaluating parallel computing systems is important to parallel systems and application design. First,based on an analysis of the current metric models for parallel computing systems,this paper classifies these model into two categories including the single computing performance model and the synthetically multiple factors productivity model. The work which has been done in the field is reviewed. Then,the various metrics are described. Something that is missing in the current research and the factors which should be considered in building the metric model are pointed out. Since the measure index for the performance of the systems demands a shift from “high performance” to “high productivity”,the paper takes the reliability and energy consumption into account,builds the reliability parallel speedup and energy parallel speedup model in view of scalability,and further proposes a metric model for evaluating the productivity of parallel computing systems. Finally,the direction of further research is proposed.

Key words: parallel computing system;metric;system evaluation