基于混合存储的无盘网络服务器数据优化分布方法  

Data Optimization Distribution Method for Diskless Network Server Based on Hybrid Storage

作　　者：谭怀亮[1] 彭诗辉[1] 贺再红[1] 

机构地区：[1]湖南大学信息科学与工程学院,长沙410082

出　　处：《计算机工程》2016年第4期7-13,共7页Computer Engineering

基　　金：国家自然科学基金资助项目(61272190)

摘　　要：无盘网络服务器需处理大量I/O操作,导致其存储子系统成为系统性能提升的瓶颈。为解决该问题,提出基于固态硬盘(SSD)和硬盘驱动器(HDD)混合存储的数据分布方法,使工作站共享的操作系统和应用程序镜像数据分布于SSD,而将工作站运行过程中的镜像修改数据分布于HDD。根据无盘网络服务器I/O数据块访问特性,建立块分布模型,由此设计SSD和HDD间的数据块动态迁移方法,根据SSD和HDD中块的访问信息计算数据分布概率,并动态决定数据块的保存位置。实验结果表明,该方法可有效提高工作站启动速度及存储系统带宽,减少SSD写次数。The storage subsystem is a bottleneck as the server has to handle a lot of I/0 operations in a diskless workstation system. In order to solve the problem, this paper presents a data distribution method based on Solid State Drive（SSD） and Hard Disk Drive （HDD） hybrid storage. It allocates OS and application image files to SSD while allocating the modified data of the image files generated by running workstations to HDD. It presents a block distribution model based on I/O features of diskless network server, and designs a dynamic migration method between SSD and HDD. The distribution method computes data distribution probability according to data access features of SSD and HDD, and then determines where a data block should be placed. Experimental results show that this method can improve the workstation boot speed,widen bandwidth of the shared storage and cut down the count of writing operations to SSD.

关 键 词：固态硬盘 硬盘驱动器 无盘网络 数据分布 块分布模型 动态块迁移 

分 类 号：TP393.04[自动化与计算机技术—计算机应用技术]