Experiences with Belinda : a synthetic Linda benchmark for parallel computer platforms

摘要: In this paper we ovewiew the design of BeLinda, present results running BeLinda on three radically different architectures, and discuss our experiences with taking particular approach to benchmarking parallel architectures. We identify some primitives that are generally used in applications, perform evaluation respect t o these attempt combine a single number. combining results, faced problems assigning suitable weights programs. These include identification typical workload, determination frequency occurrence conversion into same units. Our allows quick easy strengths weaknesses machine being evaluated, but conclude it is not realistic idea reduce overall