OpenSHMEM Active Message Extension for Task-Based Programming

OpenSHMEM Active Message Extension for Task-Based Programming
复制标题

用于基于任务的编程的 OpenSHMEM 主动消息扩展

DOI:
10.1007/978-3-031-04888-3_8
复制
发表时间:
2022
期刊:
OpenSHMEM 2021: OpenSHMEM and Related Technologies. OpenSHMEM in the Era of Exascale and Smart Networks
影响因子:
--
通讯作者:
Lu, Curtis
Lu, Curtis
中科院分区:
--
文献类型:
--
作者:
Lu, Curtis

文献摘要

参考文献

相似文献

作为一种轻量级的基于库的分区全局地址空间(PGAS)编程模型,OpenSHMEM提供了高效的单边和集体通信,近年来受到越来越多的关注。然而,基于任务的编程模型正在科学计算社区中获得更大的吸引力。面对日益增长的应用复杂性和现代高性能计算机器中日益增长的节点并行性,应用程序开发人员被它们实现更好负载平衡的能力所吸引。尽管在OpenSHMEM+X模型中,通信上下文为线程提供了对网络的一流访问,但OpenSHMEM执行其他基于任务的模型中的高级操作的能力仍然非常有限。例如,与UPC++编程模型中的远程过程调用(RPC)机制相比,如果使用信号/等待例程来实现类似的远程任务启动操作,则需要进行更多工作。在本文中,我们介绍了OpenSHMEM的一个轻量级主动消息(AM)扩展,它被设计为执行简短的、非阻塞的远程函数调用。此扩展旨在为OpenSHMEM带来基于任务的编程的一些好处,而不是使其成为具有复杂调度器的成熟的重量级任务系统。我们通过运行微基准测试程序来研究该主动消息扩展的性能,并使用TaskB边框架评估其在不同任务粒度下的计算效率。
As a lightweight library-based Partitioned Global Address Space (PGAS) programming model, OpenSHMEM provides efficient one-sided and collective communications and is receiving more attention in recent years. However, task-based programming models are getting bigger traction in scientific computing communities. Application developers are attracted by their ability to achieve better load balance in the face of ever-growing application complexity, and the increasing on-node parallelism in modern high-performance computing machines. Although communication contexts provide threads with first-class access to the network in the OpenSHMEM+X model, OpenSHMEM still has very limited ability to perform advanced operations found in other task-based models. For example, compared to the remote procedure call (RPC) mechanism in the UPC++ programming model, more work is required if the signal/wait routines are used to achieve similar remote task launching operations. In this paper, we introduce a lightweight active message (AM) extension to OpenSHMEM that is designed to perform short, non-blocking remote function invocations. This extension aims to bring some benefits of task-based programming to OpenSHMEM without making it a full-blown heavyweight tasking system with a sophisticated scheduler. We study the performance of this active message extension by running micro-benchmarks, and by evaluating its computation efficiency at different task granularities using the TaskBench framework.
DOI: 10.1109/sc41405.2020.00066
发表时间: 2019-08
期刊: SC20: International Conference for High Performance Computing, Networking, Storage and Analysis
影响因子: --
作者:
Elliott Slaughter;Wei Wu;Yuankun Fu;Legend Brandenburg;N. Garcia;Wilhem Kautz;Emily Marx;Kaleb S. Morris;Wonchan Lee;Qinglei Cao;G. Bosilca;S. Mirchandaney;Sean Treichler;P. McCormick;A. Aiken
通讯作者: Elliott Slaughter;Wei Wu;Yuankun Fu;Legend Brandenburg;N. Garcia;Wilhem Kautz;Emily Marx;Kaleb S. Morris;Wonchan Lee;Qinglei Cao;G. Bosilca;S. Mirchandaney;Sean Treichler;P. McCormick;A. Aiken
设计、实现和评估即将推出的 OpenSHMEM Teams API
DOI: --
发表时间: 2019
期刊: 2019 IEEE/ACM Parallel Applications Workshop, Alternatives To MPI (PAW-ATM)
影响因子: --
作者:
David Ozog;Md. Wasi;Gerard Taylor;James Dinan
通讯作者: James Dinan
DOI: 10.1007/s42452-020-03784-x
发表时间: 2018-06
影响因子: 2.6
作者:
Patrick Diehl;P. Jha;Hartmut Kaiser;R. Lipton;M. Lévesque
通讯作者: Patrick Diehl;P. Jha;Hartmut Kaiser;R. Lipton;M. Lévesque
GASNet-EX:适用于百亿亿次计算的高性能便携式通信库
DOI: --
发表时间: 2018
期刊: International Workshop on Languages and Compilers for Parallel Computing
影响因子: --
作者:
D. Bonachea;Paul H. Hargrove
通讯作者: Paul H. Hargrove
使用上下文在多线程 OpenSHMEM 应用程序中启用低开销通信
DOI: --
发表时间: 2019
期刊: 2019 IEEE/ACM Parallel Applications Workshop, Alternatives To MPI (PAW-ATM)
影响因子: --
作者:
Wenbin Lu;Tony Curtis;B. Chapman
通讯作者: B. Chapman