Improving the Parallel Performance of an NBody Application Using Adaptive Techniques in HPX
Zahra Khatami, Hartmut Kaiser, Jagannathan Ramanujam · 2017
The overheads of manually tuning a loop's parameters, such as chunk size, might prevent an application from reaching its maximum parallel performance. In this paper, we address this challenge by implementing a multinomial logistic regression model on an HPX loop. We present a framework that captures both the static and dynamic information of the runtime environment and feeds this information to a learning model which assigns an efficient loop chunk size automatically. Our evaluated execution results show that the proposed technique improves the performance of an NBody application by an average about 33%, 17%, and 19% compared to using existing HPX auto-parallelization tools when run with problem sizes of 105, 106, and 107particles respectively.