Evaluating a formal methodology for dynamic tuning of large-scale parallel applications

摘要

Large-scale parallel applications performance is usually far from the expected. Dynamic tuningrnis a powerful technique that helps to improve the performance of parallel applications. To bringrnthis technique to large-scale computers, this work presents a model that enables decentralizedrndynamic tuning of large-scale parallel applications. In this model, applications are decomposedrninto disjoint subsets of tasks that can be tuned individually but also abstracted to obtain a globalrnview of the parallel application. The proposed model has been designed as a hierarchical tuningrnnetwork of distributed analysis modules and implemented in the form of ELASTIC, an environmentrnfor large-scale dynamic tuning. Using ELASTIC an experimental evaluation has beenrnconducted over a synthetic large-scale parallel application and a real agent-based parallel application.rnThe results show that the proposed model, embodied in ELASTIC, is able to scale to meetrnthe demands of dynamic tuning over thousands of processes, while effectively improving thernperformance of large-scale applications.