Policies for data archival in hierarchical storage management

Abstract

We present a model to manage data across a memory hierarchy consisting of two levels (primary and secondary). The primary system (for example, a magnetic disk) is faster, but costs more per byte of storage. The secondary system (for example, an optical disk) is slower and cheaper. The data managed is assumed to exhibit time-dependent demand. A dynamic policy for transferring data from the primary system to the secondary system is derived. This policy uses information about the number of records in the primary system and transfers the least recently used (LRU) record to the secondary system, if the time elapsed since it was last used has exceeded a certain threshold. The operating characteristics of the dynamic policy are studied both analytically and numerically. The dynamic policy is compared to two static policies in a numerical experiment. A static policy is one that does not utilize information about the number of records in the primary system. Our results show that the dynamic policy outperforms the two static policies in a variety of situations. We also study several policies that lie in between dynamic and static policies.