MANA for MPI: MPI-Agnostic Network-Agnostic Transparent Checkpointing

MANA for MPI: MPI-Agnostic Network-Agnostic Transparent Checkpointing
复制标题

MANA for MPI:与 MPI 无关的网络与透明检查点

DOI:
10.1145/3307681.3325962
复制
发表时间:
2019
期刊:
Proc. of 28th Int. Symp. on High Performance Parallel and Distributed Computing (HPDC'19
影响因子:
--
通讯作者:
Cooperman, Gene
Cooperman, Gene
中科院分区:
--
文献类型:
--
作者:
Garg, Rohan;Price, Gregory;Cooperman, Gene

文献摘要

参考文献

被引文献

相似文献

在高性能计算中,并行检查点MPI的容错和负载平衡是一个长期存在的问题。由于需要在所有网络互连上为MPI实现的所有组合提供检查点重启服务,因此问题变得复杂。这项工作提出了MANA(MPI无关的网络无关的透明检查点),一个单一的代码库,支持所有的MPI实现和互连组合。不可知属性意味着可以在一个MPI实现下(可能通过TCP)对MPI应用程序进行检查点,然后在每个节点具有不同数量CPU内核的群集上通过InfiniBand在第二个MPI实现下重新启动。这种技术是基于一种新的“分裂进程”的方法,它使两个独立的程序共存于一个单一的进程与一个单一的地址空间。这项工作克服了两个最广泛采用的透明检查点解决方案,BLCR和DMTCP/InfiniBand,需要单独修改每个MPI实现和/或底层网络API的限制。运行时开销被发现是微不足道的检查点重新启动在一个单一的主机,当比较一个本地MPI计算迁移到一个远程集群对一个普通的MPI计算本地运行在同一个远程集群。
Transparently checkpointing MPI for fault tolerance and load balancing is a long-standing problem in HPC. The problem has been complicated by the need to provide checkpoint-restart services for all combinations of an MPI implementation over all network interconnects. This work presents MANA (MPI-Agnostic Network-Agnostic transparent checkpointing), a single code base which supports all MPI implementation and interconnect combinations. The agnostic properties imply that one can checkpoint an MPI application under one MPI implementation and perhaps over TCP, and then restart under a second MPI implementation over InfiniBand on a cluster with a different number of CPU cores per node. This technique is based on a novel "split-process" approach, which enables two separate programs to co-exist within a single process with a single address space. This work overcomes the limitations of the two most widely adopted transparent checkpointing solutions, BLCR and DMTCP/InfiniBand, which require separate modifications to each MPI implementation and/or underlying network API. The runtime overhead is found to be insignificant both for checkpoint-restart within a single host, and when comparing a local MPI computation that was migrated to a remote cluster against an ordinary MPI computation running natively on that same remote cluster.
DOI: 10.1109/cluster.2018.00047
发表时间: 2018-08
期刊: 2018 IEEE International Conference on Cluster Computing (CLUSTER)
影响因子: --
作者:
Rohan Garg;Apoorve Mohan;Michael B. Sullivan;G. Cooperman
通讯作者: Rohan Garg;Apoorve Mohan;Michael B. Sullivan;G. Cooperman
使用 MPI 工具接口进行 MPI 性能工程:MVAPICH 和 TAU 的集成
DOI: --
发表时间: 2017
期刊: EuroMPI/USA
影响因子: --
作者:
Srinivasan Ramesh;Aurèle Mahéo;S. Shende;A. Malony;H. Subramoni;D. Panda
通讯作者: D. Panda
MPI 阶段:批量同步应用程序的 MPI 状态检查点
DOI: 10.1145/3236367.3236385
发表时间: 2018
期刊: Proceedings of the 25th European MPI Users' Group Meeting
影响因子: --
作者:
Nawrin Sultana;A. Skjellum;I. Laguna;M. Farmer;K. Mohror;M. Emani
通讯作者: M. Emani
使用进程级虚拟化的分布式资源检查点的设计和实现
DOI: 10.1109/cluster.2016.55
发表时间: 2016
期刊: 2016 IEEE International Conference on Cluster Computing (CLUSTER)
影响因子: --
作者:
K. Arya;Rohan Garg;A. Y. Polyakov;G. Cooperman
通讯作者: G. Cooperman
进程劫持
DOI: 10.1109/hpdc.1999.805296
发表时间: 1999
期刊: Proceedings. The Eighth International Symposium on High Performance Distributed Computing (Cat. No.99TH8469)
影响因子: --
作者:
Victor C. Zandy;Barton P. Miller;Miron Livny
通讯作者: Miron Livny