On Accurate Performance Prediction for Massively Parallel Systems and its Applications

A performance prediction method is presented, which accurately predicts the execution time of programs on massively parallel computers. We consider a computer architecture with shared-memory multiprocessor nodes and a communication network connecting these nodes. The method combines a relaxed task graph model, a queueing model and a memory hierarchy model. The relaxed task graph is a compact representation of communicating processes of an application mapped on the target machine. Simultaneous accesses to the resources of a multiprocessor node are modeled as a queueing network. The execution time of the application is computed by an evaluation algorithm. An example application implemented on a massively parallel computer demonstrates the accuracy of our model. Furthermore two applications of our accurate prediction method are presented.