Task priority control for the HPX runtime system

Suhang Jiang, Mulya Agung, Ryusuke Egawa, Hiroyuki Takizawa

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

High Performance ParalleX (HPX) is one of taskbased programming and execution models, which provides a C++ class library to describe tasks and their dependencies, and also a runtime system for task-based distributed-memory parallel computing. In the original HPX mechanism, the execution of some critical tasks might be delayed by executing other noncritical tasks. This paper hence proposes to incorporate task priority control into the HPX runtime system. First, this paper discusses use of multiple thread pools to prioritize critical tasks. The waiting time of critical tasks is reduced by assigning critical tasks to a dedicated thread pool that is different from the thread pool for non-critical tasks. Second, this paper discusses how to assign worker threads to processor cores. A different number of worker threads are associated with each thread pool, and then mapped to processor cores so as to mitigate the load imbalance among the NUMA domains. In this paper, the performance of the proposed mechanism is evaluated using two task-based applications and three different scales of NUMA systems. The evaluation results clearly demonstrate that the proposed mechanism can improve the performance of HPX applications by successfully prioritizing critical tasks and also reducing the load imbalance of the NUMA domains.

Original languageEnglish
Title of host publicationProceedings - 2020 IEEE 34th International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages806-813
Number of pages8
ISBN (Electronic)9781728174457
DOIs
Publication statusPublished - 2020 May
Event34th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2020 - New Orleans, United States
Duration: 2020 May 182020 May 22

Publication series

NameProceedings - 2020 IEEE 34th International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2020

Conference

Conference34th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2020
CountryUnited States
CityNew Orleans
Period20/5/1820/5/22

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Hardware and Architecture
  • Safety, Risk, Reliability and Quality
  • Control and Optimization

Fingerprint Dive into the research topics of 'Task priority control for the HPX runtime system'. Together they form a unique fingerprint.

Cite this