THCAS:内容存储系统及其性能优化  

THCAS:Content-aware storage system and performance optimization

作　　者：宋铭[1] 刘川意[1] 张悠慧[1,2] 汪东升[1,2] 

机构地区：[1]清华大学计算机科学与技术系,北京100084 [2]清华大学微处理器与片上系统技术研究中心,北京100084

出　　处：《清华大学学报（自然科学版）》2009年第7期1032-1036,共5页Journal of Tsinghua University(Science and Technology)

基　　金：国家"八六三"高技术项目(2006AA01Z111);国家"九七三"重点基础研究计划项目(2007CB310900)

摘　　要：为实现开源的基于内容存储系统及性能优化,以应对固定内容数据存储的急速增长而带来的对这种存储系统结构的广泛需求,该文设计并实现了清华大学的基于内容存储(THCAS)系统,并对系统性能进行了分析和评价,在此基础上,提出了多线程Hash和存储流水线2种基于内容存储的优化方法。在正常操作系统I/O Buffer和去掉异步I/O 2种情况下,将优化后的系统与原系统的写数据吞吐率和CPU使用率进行测试比较,优化后系统在2种情况下均提升了原系统的写数据性能,多线程Hash提高了150%,而存储流水线提高了25%。The rapid increase in the amount of fixed content data is creating more needs for content-aware storage which should be designed based on an open-source implementation. This paper describes a Tsinghua University designed content-aware storage system, THCAS, that uses multi-thread Hashes and a storage pipeline to optimize the system performance. The file-writing throughput and the CPU utilization for an optimized system with and without an I/O buffer are compared with the original system. Tests demonstrate that the optimized content-aware storage with the multi-thread Hash and the storage pipeline significantly improve the performance with the multi-thread Hash improveing the write performance by 150% while the storage pipeline improves it by 25%.

关 键 词：基于内容存储 HASH 存储流水线 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]