Mobile computing suffers from both uncertain connectivity and limited battery life. Data grouping and predictive prefetching can help in both areas. When communicating over an unreliable link, mobile computer users risk not having access to needed data. The manual hoarding of files both puts considerable burden on the user and is likely to be inaccurate or incomplete due to limited knowledge. By identifying groups of highly related files through past access patterns, the system can preemptively hoard files on a mobile client thus reducing the client's dependence on an unreliable or high-latency communications links, while making better use of limited network bandwidth. Careful prefetching of data will transform the disk workloads which should increase the available number and duration of idle periods where the disk can be spun down. File access events are typically spread over a period of time while intervening processing takes place. Previously observed access patterns can enable the prefetching of a series of files, thus localizing the accesses to a short period. This creates longer disk idle periods where the disk can be spun down.

We are currently focusing on grouping and prefetching algorithms for power conservation in mobile computers. This involves the analysis of file system traces and measuring the effects of various prefetching policies.