Efficient and Accurate Selection of Optimal Collective Communication Algorithms Using Analytical Performance Modeling

Abstract

The performance of collective operations has been a critical issue since the advent of Message Passing Interface (MPI). Many algorithms have been proposed for each MPI collective operation but none of them proved optimal in all situations. Different algorithms demonstrate superior performance depending on the platform, the message size, the number of processes, etc. MPI implementations perform the selection of the collective algorithm empirically, executing a simple runtime decision function. While efficient, this approach does not guarantee the optimal selection. As a more accurate but equally efficient alternative, the use of analytical performance models of collective algorithms for the selection process was proposed and studied. Unfortunately, the previous attempts in this direction have not been successful. We revisit the analytical model-based approach and propose two innovations that significantly improve the selective accuracy of analytical models: (1) We derive analytical models from the code implementing the algorithms rather than from their high-level mathematical definitions. This results in more detailed models.(2) We estimate model parameters separately for each collective algorithm and include the execution of this algorithm in the corresponding communication experiment. We experimentally demonstrate the accuracy and efficiency of our approach using Open MPI broadcast and gather algorithms and two different Grid'5000 clusters and one supercomputer.

Algoritmo de Árvore Binomial Balanceada para Broadcast com MPI

“…Segundo [Nuriyev and Lastovetsky 2021], embora a func ¸ão de selec ¸ão de algoritmos do MPI seja muito eficiente, ela não pode garantir a melhor opc ¸ão para todas as situac ¸ões. Uma maior análise de diversas situac ¸ões diferentes seria interessante para que seus resultados fossem usados para criar uma nova func ¸ão mais adequada para a escolha de algoritmos de comunicac ¸ão, provavelmente com mais parâmetros para chegar em conclusões mais precisas, mas simples o suficiente para ainda ser eficiente.…”

“…O Open MPI utiliza segmentac ¸ão de mensagens [Nuriyev and Lastovetsky 2021] nos algoritmos de broadcast, com excec ¸ão do algoritmo Linear Tree. O trabalho também aborda a implementac ¸ão do algoritmo Binomial Tree (BT) no MPI, que é feita usando uma combinac ¸ão de Linear Trees com sends e receives não-bloqueantes.…”