arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1906.05020cs.DC
收藏

基于线程的MPI运行时的检查点/重启方法

Checkpoint/restart approaches for a thread-based MPI runtime

Julien Adam, Maxime Kermarquer, Jean-Baptiste Besnard, Leonardo Bautista-Gomez, Marc Perache, Patrick Carribault, Julien Jaeger, Allen D. Malony, Sameer Shende

摘要:

在大规模运行大规模并行程序时,容错一直是一个重要的话题。从统计数据来看,硬件和软件故障预计会更频繁地发生在收集数百万计算单元的系统上。此外,作业越大,崩溃浪费的计算时间就越多。在本文中,我们描述了我们的MPI运行时所做的工作,使透明和应用程序级的检查点机制。与MPI 4.0用户级故障缓解(ULFM)接口不同,我们的工作仅针对检查点/重启,而忽略了其他功能,如弹性。我们展示了如何现有的检查点方法可以实际应用到基于线程的MPI实现足够的运行时协作。两个主要贡献是在透明C/R期间保持高速网络性能,以及由于专用的用户级调度器支持而超额订阅检查点数据复制。这些技术在MPI基准测试(如IMB,Lulesh和Heatdis)上进行了测量,并讨论了相关的开销和权衡。

英文摘要:

Fault-tolerance has always been an important topic when it comes to running massively parallel programs at scale. Statistically, hardware and software failures are expected to occur more often on systems gathering millions of computing units. Moreover, the larger jobs are, the more computing hours would be wasted by a crash. In this paper, we describe the work done in our MPI runtime to enable both transparent and application-level checkpointing mechanisms. Unlike the MPI 4.0 User-Level Failure Mitigation (ULFM) interface, our work targets solely Checkpoint/Restart and ignores other features such as resiliency. We show how existing checkpointing methods can be practically applied to a thread-based MPI implementation given sufficient runtime collaboration. The two main contributions are the preservation of high-speed network performance during transparent C/R and the over-subscription of checkpoint data replication thanks to a dedicated user-level scheduler support. These techniques are measured on MPI benchmarks such as IMB, Lulesh and Heatdis, and associated overhead and trade-offs are discussed.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑