基于SANs模型的一种并行I/O系统的可用性评估

计算机工程与应用 ›› 2008, Vol. 44 ›› Issue (19): 67-71.

基于SANs模型的一种并行I/O系统的可用性评估

郑霄^1,2,李宏亮²,郑方²,郑翔²,陈左宁²

1.解放军信息工程大学，郑州 450002
2.江南计算技术研究所，江苏无锡 214083

收稿日期:2008-03-06 修回日期:2008-04-16 出版日期:2008-07-01 发布日期:2008-07-01
通讯作者: 郑霄

Availability evaluation based on SANs model for a parallel I/O system

ZHENG Xiao^1,2,LI Hong-liang²,ZHENG Fang²,ZHENG Xiang²,CHEN Zuo-ning²

1.PLA Information and Engineering University，Zhengzhou 450002，China
2.Jiangnan Institute of Computing Technology，Wuxi，Jiangsu 214083，China

Received:2008-03-06 Revised:2008-04-16 Online:2008-07-01 Published:2008-07-01
Contact: ZHENG Xiao

摘要/Abstract

摘要： 并行I/O系统是高性能计算机系统的一个重要组成部分，其可用性水平对整机系统性能的发挥具有重要作用。采用SANs（Stochastic Activity Networks，随机行为网）模型及其支持工具Mobius，对一种大规模并行I/O系统建立可用性评估模型，并采用模拟方法进行解析。模拟结果反映了全局文件系统数量、单一文件系统内最小可用OST（Object Storage Target，对象存储目标）数量和系统维修时间等参数的变化对全系统可用度的影响，对于大规模并行I/O系统的设计与维护具有积极的参考价值。

关键词: 随机行为网, Mobius, 并行I/O系统, 可用性评估

Abstract: The parallel I/O system is one of important components of a HPC system，and it can discount the performance of the whole system if its availability are awful.However，high-throughput and large capacity are still the main target in parallel I/O system’s design by far，and few research on its availability evaluation has been published.In this paper，a method managing to evaluate a parallel I/O system’s availability based on SANs（Stochastic Activity Networks） model with Mobius modeling environment is expatiated，and the simulation results reveal how the availability of the parallel I/O system is impacted on by the number of global file system，the number of minimum number of OST in a single file system，and the repair period of the whole system.

Key words: Stochastic Activity Networks（SANs）, Mobius, parallel I/O system, availability evaluation

郑霄^1,2,李宏亮²,郑方²,郑翔²,陈左宁². 基于SANs模型的一种并行I/O系统的可用性评估[J]. 计算机工程与应用, 2008, 44(19): 67-71.

ZHENG Xiao^1,2,LI Hong-liang²,ZHENG Fang²,ZHENG Xiang²,CHEN Zuo-ning²

. Availability evaluation based on SANs model for a parallel I/O system[J]. Computer Engineering and Applications, 2008, 44(19): 67-71.