...
首页> 外文期刊>Parallel Processing Letters >MASSIVELY PARALLEL I/O FOR PARTITIONED SOLVER SYSTEMS
【24h】

MASSIVELY PARALLEL I/O FOR PARTITIONED SOLVER SYSTEMS

机译:分区求解器系统的大规模并行I / O

获取原文
获取原文并翻译 | 示例
           

摘要

This paper investigates I/O approaches for massively parallel partitioned solver systems.nTypically, such systems have synchronized “loops” and write data in a well defined blocknI/O format consisting of a header and data portion. Our target use for such a parallel I/Onsubsystem is checkpoint-restart where writing is by far the most common operation andnreading typically only happens during either initialization or during a restart operationnbecause of a system failure. We compare four parallel I/O strategies: POSIX File PernProcessor (1PFPP), “Poor-Man’s” Parallel I/O (PMPIO), a synchronized parallel I/On(syncIO), and a “reduced blocking” strategy (rbIO). Performance tests executed on thenBlue Gene/P at Argonne National Laboratory using real CFD solver data from PHASTAn(an unstructured grid finite element Navier-Stokes solver) show that the syncIO strategyncan achieve a read bandwidth of 47.4 GB/sec and a write bandwidth of 27.5 GB/secnusing 128K processors. The “reduced-blocking” rbIO strategy achieves an actual writingnperformance of 17.8 GB/sec and the perceived writing performance is 166 TB/sec onnBlue Gene/P using 128K processors.
机译:本文研究了大规模并行分区求解器系统的I / O方法。通常,此类系统具有同步的“循环”,并以定义明确的blockI / O格式写入数据,该格式由标头和数据部分组成。这种并行I / OnSubsystem的目标用途是检查点重新启动,其中写入是迄今为止最常见的操作,并且由于系统故障,通常仅在初始化或重新启动操作期间执行读操作。我们比较了四种并行I / O策略:POSIX文件PernProcessor(1PFPP),“穷人”并行I / O(PMPIO),同步并行I / On(syncIO)和“减少阻塞”策略(rbIO)。使用来自PHASTAn(非结构化网格有限元Navier-Stokes求解器)的真实CFD求解器数据在Argonne国家实验室的thenBlue Gene / P上执行的性能测试表明,syncIO策略可以实现47.4 GB /秒的读取带宽和27.5的写入带宽GB /正在使用128K处理器。 “减少阻止” rbIO策略在使用128K处理器的nBlue Gene / P上实现了17.8 GB /秒的实际写入性能,感知的写入性能为166 TB /秒。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号