An Approach for Migrating Data Adaptively in Hierarchical Storage Systems

Hierarchical storage management (HSM) constructs a storage system with different types of devices, e.g., SCSI disks, SATA disks, FC disks and even SSD disks. The goal is to obtain large capacity with low cost. In order to achieve high performance, hierarchical storage management systems classify data dynamically and move them between fast devices and slow devices efficiently. However, almost all existing HSM systems have some limitations. Especially, without taking full advantage of the information from workloads, they make data migration affect application performance noticeably. AutoMig, an approach to automatically migrate data in a HSM system is proposed to enhance I/O performance of foreground applications. Firstly, AutoMig dynamically classifies files according to file access history, file size and storage utilization. Furthermore, an LRU queue is used to maintain the files in the fast devices. Secondly, it finds correlated files for automatic prefetching with data mining technology. Thirdly, AutoMig applies different policies to files migrating actions. It adaptively adjusts the migration rate according to varying workloads for migrating out actions while uses the asfastaspossible policy for migrating back actions. The application in a hierarchical storage system showes that AutoMig can effectively shorten foreground I/O response time compared with existing approaches.