并行可视化调试工具中检查点使能的研究与实现.pdfVIP

第 22 卷第 10 期 计算机应用与软件 Vol.22, No.10 2005 年 10 月 Computer Applications and Software Oct. 2005 并行可视化调试工具中检查点使能的研究与实现 臧德华 邓倩妮 陆鑫达 (上海交通大学计算机科学与工程系网格中心 上海 200030) 摘 要 与串行程序相比，并行程序调试会遇到新的问题。首先并行程序往往需要长时间运行，从而导致并行程序调试是一个 尤其费时的过程；其次并行程序调试过程中，某一次调试出现的错误在下次调试的时候不一定出现，给错误跟踪带来了很大困难。 本文针对这两个问题，设计和实现了一个中间件系统，在并行调试工具 XMPI 中使能BLCR 检查点系统的。通过该中间件，在使 用 XMPI 调试大型 MPI 并行程序的时候，减少调试阶段并行程序运行时间，并且可以更好跟踪并行程序错误，提高并行程序开发 效率。 关键词 BLCR LAM-MPI XMPI 使能中间件 检查点 重启 THE RESEACH AND IMPLEMENTATION OF ENABLING CHECKPOINT IN VISUAL PARALLEL DEBUGGING TOOLS Zang Dehua Deng Qianni Lu Xinda (Grid Computing Center, Dept. of Computer Science and Engineering, Shanghai Jiaotong University, Shanghai 200030, China) Abstract Compared with serial programs, parallel programs debugging will face some new problems. Firstly during the parallel program debugging, they are usually needed to run a long time to find the bugs, which is time wasteful. Secondly the problems occurring in a debugging maybe won’t happen in the next time, which leads to great difficulty to track the problems. This paper focuses on these two problems, and designs and implements an enabling component which enables the BLCR checkpoint and restart system in the parallel debugging tool XMPI. During debugging MPI parallel programs in XMPI, the debugging time is cut shortly and the faults can be tracked, and the efficient of developing parallel programs is improved. Keywords BLCR LAM-MPI XMPI Enabling component Checkpoint Restart XMPI 。第三部分重点论述了检查点使能中间件系统的设计和实 1 引 言 现。第四部分给出并行可视化