基于块的分级存储系统多样性机制设计与研究  

Diversity Mechanisms Design and Research of Block-based Hierarchical Storage System

作　　者：杜然[1,2] 黄秋兰[1] 程耀东[1] 陈刚[1] 

机构地区：[1]中国科学院高能物理研究所,北京100049 [2]中国科学院大学,北京100049

出　　处：《计算机工程》2016年第12期50-59,共10页Computer Engineering

基　　金：国家自然科学基金(11305192;11575223;11375221;11405195;11205177);中国科学院"十二五"科研信息化项目"高能物理科技领域云"(Y2293C00S5)

摘　　要：HazelNut是一种基于块的分级存储系统,为满足HazelNut多种存储设备和追踪迁移方式的管理需求,设计并实现多样性设备管理机制HNDisk和多样性追踪迁移机制HNType。HNDisk将物理存储设备分层管理,不限制设备种类和设备数量。HNType可按照公共接口添加类型模块,便于配置并选择不同追踪度量和迁移算法。在HNDisk和HNType的基础上,测试不同追踪度量对btier平均值迁移算法性能的影响,结果表明,追踪度量访问次数和访问字节数非线性相关,两者对迁移效果的影响不同,并且使用读写请求完成时间作为追踪度量至少可使顺序读写性能提升10%。HazelNut is a block-based hierarchical storage system, which requires to manage multiple types of storage devices and different ways of I/O trace and block migration. As a result, a diversity equipment management mechanism （HNDisk） and diversity trace migration mechanism（HNType） are designed and implemented to meet these requirements. HNDisk is able to decentralized organize multiple storage devices in tiers with no constraints on device types and device amount. Meanwhile,HNType can add the type module according to the common interface,and it is easy to configure and select different trace metrics and migration algorithm. At the same time, it tests the influence of different trace metrics on the performance of btier average migration algorithm on the base of HNDisk and HNType. Experimental result show that trace metrics of access times and access sectors are not linearly correlated,as a result,these two metrics bring different migration results. Besides,I/O requests completion time used as trace metrics is able to improve sequential I/O by at least 10%.

关 键 词：分级存储系统 多样性 LINUX内核 可装载内核模块 虚拟文件系统开关 追踪度量 迁移算法 

分 类 号：TP316.4[自动化与计算机技术—计算机软件与理论]