Performance Prediction of Dynamic Task Structures with N-map

A set of computerized tools to support a performance oriented parallel program development is integrated in the CAPSE (Computer Aided Parallel Software Engineering) environment. The respective tool support for performance engineering activities covers performance prediction in the design phase, analytical or simulation based performance evaluation in the detailed speciication and coding phase, and monitoring in the test phase. The N-MAP tool, as part of the CAPSE environment and dedicated for early performance prediction, intends an incremental development process of parallel applications such that implementation design choices can be investigated far ahead of the full coding of the application. N-MAP, from a skeletal representation of a parallel SPMD program that covers just those program structures that have critical performance impacts, in a compilation step automatically generates a discrete event simulation program, the execution of which exactly reeects (and generates) the behavior of the skeletal parallel program. Various performance related investigations can be conducted based on the trace statistics coming from the simulated execution of the skeleton. N-MAP has been successfully used for the development of numerical applications with a static structure of task occurrences. In this work we demonstrate how inherently dynamic task communication structures can be analyzed using N-MAP, by incrementally developing the implementation of a general purpose distributed discrete event simulation engine based on the Time Warp protocol. The dynamicity of the behavior of Time Warp is known to pose a stress case for any performance prediction methodology. We show how dynamic behavior is characterized in terms of probabilis-tic workload parameters, based on which N-MAP can generate adequate performance predictions.