ISRA-Based Grouping: A Disk Reorganization Approach for Disk Energy Conservation and Disk Performance Enhancement

摘要

Reducing disk energy consumption and improving disk performance in high-performance computer systems are increasingly pressing issues for reasons of disk economy and efficiency. To achieve these goals, we define the concept of Immediate Successor Relationship Amount (ISRA) to represent the successor relationship of data blocks, and propose an ISRA-based grouping algorithm for disk reorganization, based on an undirected graph. We group data blocks that experience frequent successive accesses, then sort them using a merge-sort-like algorithm to determine the position of every group as well as the new position of every block within those groups. We evaluate our approach in terms of disk seek time and disk energy consumption, using Disksim and the log energy model. The results show clearly that both disk seek time and the energy needs can be reduced by about 50 percent.