Cost-Effective Jukebox Storage via Hybrid File-Block Caching
NGIT '99 Proceedings of the 4th International Workshop on Next Generation Information Technologies and Systems
Analysis of Workload Behavior in Scientific and Historical Long-Term Data Repositories
ACM Transactions on Storage (TOS)
| Hi-index | 0.00 |
A study of file access patterns can be useful in designing an efficient hierarchical storage system that employs automatic file migration. This paper proposes a specific design for such a storage system, and describes a detailed study of long-term file access patterns on a number of UNIX file systems in use at Berkeley. File access traces are collected and used to drive a file migration system simulator. The results of the study are used to drive a file migration system simulator. The results of the study are used to support the argument that the proposed storage system design is reasonable, and to propose file migration algorithms that might best suit the storage system. Previous studies of this topic took similar approaches. However, they were based on data collected ten or more years ago from systems designed very differently than those of today. Also, because the storage capacity of memory and disks has increased 50 to 100 times in the past ten years, there is good reason to question whether the results of previous studies are valid for today''s systems. Although the distributed UNIX- based system we trace in this study is vastly different that the large time-sharing systems used in previous studies, we find that some of the same types of migration algorithms still perform well. Specifically, file replacement algorithms based on multiplying the file size by the time since last access to obtain an ordering of migration preference between files works well.