SCALA: A Performance System For Scalable Computing

Lack of effective performance-evaluation environments is a major barrier to the broader use of high performance computing. Conventional performance environments are based on profiling and event instrumentation. It becomes problematic as parallel systems scale to hundreds of nodes and beyond. A framework of developing an integrated performance modeling and prediction system, SCALability Analyzer (SCALA), is presented in this study. In contrast to existing performance tools, the program performance model generated by SCALA is based on scalability analysis. SCALA assumes the availability of modern compiler technology, adopts statistical and symbolic methodologies, and has the support of browser interface. These technologies, together with a new approach of scalability analysis, enable SCALA to provide the user with a more intuitive level of performance analysis for scalable computing. A prototype SCALA system has been implemented. Initial experimental results show that SCALA is unique in its ability of revealing the scaling properties of a computing system.