Simplifying Automated Hoarding Methods

摘要

A number of mobile computing systems have used the technique of hoarding, which allows a mobile device to store a chosen subset of known files, to give disconnected users the illusion of a complete filesystem in the presence of limited storage. We undertook an extensive and detailed simulation study of the parameters of a well-known hoarding system, SEER, in an attempt to discover the parameters that would produce the best performance. To our surprise, we discovered that the best parameter combinations were those that completely disabled SEER'S complex clustering methods, reverting the system instead to a modified form of LRU hoarding. We discuss the experiments and our results, and propose designs for future systems and directions for future research.