Because of the limited storage space available on portable computers, disconnected mobile users must restrict their work to a subset of the files available on their network. The list of files needed to accomplish useful work is large, non-intuitive, and constantly changing. Selecting a subset by hand is difficult, time-consuming, and error-prone, suggesting that an automated solution is desirable.
Our thesis is that it is possible and practical to automate the process of choosing files to be stored on a portable computer. To validate this thesis, we conducted a preliminary study in a live business environment, which demonstrated that the approach was feasible.
We then developed a new metric, semantic distance, that quantifies the relationships among files, so that the group of files needed to work on a particular project can be identified. Using this metric, we built an automated system named SEER, which dynamically analyzes user behavior to identify the files needed for various projects, predicts the projects on which the user will be working, and then arranges to store the files necessary for these projects on the portable computer.
After building the system, we developed new metrics to characterize the behavior of automated hoarding systems, and deployed SEER among a small group of users. To our knowledge, ours is the first quantitative study of a hoarding system that has been done anywhere. The results of the study showed that SEER performed superbly, usually requiring only about a third of the hoard space needed by previous algorithms, and generally performing within a few percent of optimality. In live usage, SEER nearly always hoards 100% of the files needed by the user.
Postscript of the dissertation is available either in the original single-sided, double-spaced form (301 pages total) or paper-saving double-sided, single-spaced form (174 pages) as Technical Report UCLA-CSD-970015.
This page maintained by Geoff Kuenning.