Date of Original Version



Technical Report

Rights Management

All Rights Reserved

Abstract or Description

Abstract: "In many computer systems today, users don't have enough disk space and would welcome a method that made more efficient use of disk. It has long been hoped that compression could be used to increase the ammount of user data stored per gigabyte of disk; thus far, attempts to use compression have enjoyed only limited success due to performance degradation, increased hardware cost, or difficulty of implementation. We present and evaluate a method of using compression that does not suffer from the usual drawbacks. The idea is to have the filesystem separate disk storage into two levels of the memory hierarchy. Files that have not been used recently are compressed, while other files are kept in standard uncompressed form.In this way, the concepts of caching and compression are used to split a single disk into two levels of storage with different cost and performance characteristics. Inactive files are stored more cheaply, but take longer to access. We call this the ATTIC method. It is shown that this approach yields an increase in storage capacity nearly as great as that which could be achieved by compressing every file, with nearly the speed of a filesystem that does no compression. Data is presented showing the long term temporal locality of file accesses for 9 systems from 3 different sites. This locality is what makes file migration practical. An analysis of the performance of compression algorithms is presented.Data compressibility and file access patterns vary from machine to machine, but for many systems, the ATTIC approach would result in twice the effective storage capacity while averaging less than 10 seconds of additional delay per user per day. ATTIC has significant advantages over previous methods. It results in much greater storage capacity than a standard filesystem. It does not require any extra hardware. Once implemented, it can be easily added to existing systems. The compression of files is invisible to users, except for the increased storage and a slight degradation in performance."