Jozo J. Dujmovic San Francisco State University
We present metrics for computing differences between individual benchmarks in a benchmark suite. The minimum difference is 0 (for identical benchmarks) and the maximum difference is 1 (or 100%, for the most different benchmarks). Individual benchmarks can be geometrically interpreted as points in a "program space". Redundant benchmarks are visualized as clusters of close points.
Using quantitative differences between individual benchmarks and cluster analysis techniques it is possible to develop efficient and practical quantitative methods for evaluation and design of benchmark suites. The main design goals are to eliminate excessive redundancy and to achieve a desired distribution of benchmarks in the program space. In particular, we propose the design of universal benchmark suites which have the following properties: (1) uniform and low redundancy between component workloads, (2) maximum size, and (3) uniform distribution of workloads within the program space.
The proposed approach to standard performance evaluation has the following main advantages:
(1) For each benchmark suite it is possible to provide a scientific proof of the validity of selecting component benchmarks.
(2) The number of component benchmarks can be minimized.
(3) Standard performance indicators can be customized and made workload-sensitive.
(4) The process of updating benchmark suites can be strictly controlled, less frequent, faster, and less expensive. These features can increase the credibility and versatility of standard industrial benchmarks, and significantly reduce the cost of benchmarking.