Efficient dynamic pinning of parallelized applications by reinforcement learning with applications
Abstract
This paper describes a dynamic framework for mapping the threads of parallel applications to the computation cores of parallel systems. We propose a feedback-based mechanism where the performance of each thread is collected and used to drive the reinforcement-learning policy of assigning affinities of threads to CPU cores. The proposed framework is flexible enough to address different optimization criteria, such as maximum processing speed and minimum speed variance among threads. We evaluate the framework on the Ant Colony optimization parallel benchmark from the heuristic optimization application domain, and demonstrate that we can achieve an improvement of 12% in the execution time compared to the default operating system scheduling/mapping of threads under varying availability of resources (e.g. when multiple applications are running on the same system).
Citation
Chasparis , G , Rossbory , M & Janjic , V 2017 , Efficient dynamic pinning of parallelized applications by reinforcement learning with applications . in F F Rivera , T F Pena & J C Cabaleiro (eds) , Euro-Par 2017: Parallel Processing : 23rd International Conference on Parallel and Distributed Computing, Santiago de Compostela, Spain, August 28 – September 1, 2017, Proceedings . Lecture Notes in Computer Science (Theoretical Computer Science and General Issues) , vol. 10417 , Springer , Cham , pp. 164-176 , 23rd International Conference on Parallel and Distributed Computing (Euro-Par) , Santiago de Compostela , Spain , 28/08/17 . https://doi.org/10.1007/978-3-319-64203-1_12 conference
Publication
Euro-Par 2017: Parallel Processing
ISSN
0302-9743Type
Conference item
Description
Funding: This work has been partially supported by the European Union grant EU H2020-ICT-2014-1 project RePhrase (No. 644235).Collections
Items in the St Andrews Research Repository are protected by copyright, with all rights reserved, unless otherwise indicated.