A PERFORMANCE ADVISER FOR THE DEVELOPMENT OF PARALLEL PROGRAMS

The increasing complexity of parallel computing systems has brought ab out a crisis in parallel performance evaluation and tuning. Tools for performance measurement and visualization become necessary parts of pr ogramming environments for parallel computers. In this paper we descri be a tool - which we call the Performance Adviser - that offers two di fferent levels of performance information (high level and low level), guides the users to specific problem areas in the source code, and sug gests actions to improve the performance of their parallel programs. W orking behind the Performance Adviser is an expert system which derive s high level concepts from the source code and a critical path analysi s metric which derives low level performance information from the perf ormance data collected in the execution of the program.