一种识别I/O Feature的文件预测模型  

File Prediction Model Identifying I/O Feature

作　　者：杨晓芬[1] 师明[1] 李博[1] 胡海燕[1] 刘轶[1] 

机构地区：[1]北京航空航天大学计算机学院中德联合软件研究所,北京100191

出　　处：《小型微型计算机系统》2012年第11期2493-2497,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61073011)资助;国家"八六三"高技术研究发展计划项目(2009AA01A129)资助

摘　　要：在文件预取技术中,如何提高文件预取的命中率和适用度一直是研究的焦点.尤其是在面对大批量数据读取的时候,如何提高预取命中率对系统的性能提升有着至关重要的影响.本文提出识别I/O Feature的预测模型(IOPM),该模型通过记录文件的历史访问信息获取I/O Features,然后分析这些I/O访问模式,设计一个简单高效的特征符号表来表示这些模式.此预测模型可以有效地识别出顺序读、固定点读、逆序读、跳读、多步跳读等多种模式.同时,该模型添加应用程序的信息,可以有效地对不同程序之间的交叉读做出预测,具有很高的预测命中率.During existing I/O prefetching techniques,how to improve hit ratio and applicability of I/O prefetching has been the focus of research.Especially in the face of large-scale-data-reading,higher hit ratio is one of key factor for the system performance.The paper proposes a prediction model identifying I/O feature which extracts I/O Features from the collected file access information,analyzes the I/O access patterns and updates them automatically.This model can effectively identify sequential-read,fix-read,read-backwards,stride-read and so on.At the same time,this model can predict mix-read between different applications effectively because it has added the information of application so that it has higher hit ratio of prediction.

关 键 词：文件预取 命中率 预测模型 I O FEATURE 

分 类 号：TP303[自动化与计算机技术—计算机系统结构]