PAS2P Tool, Parallel Application Signature for Performance Prediction

摘要

Accurate prediction of parallel applications' performance is becoming increasingly complex. We seek to characterize the behavior of message-passing applications by extracting a signature to predict the performance in different target systems. We have developed a tool we called Parallel Application Signature for Performance Prediction (PAS2P) that strives to describe an application based on its behavior. Based on the application's message-passing activity, we have been able to identify and extract representative phases, with which we created a signature. We have experimented using scientific applications and we predicted the execution times on multicore architectures with an average accuracy of over 97%.