CDA 95-29304 Kaiser, Gail E. Duchamp, Daniel J. Columbia University Semantics-based Prefetching for Mobile Computing This proposal investigates collaborative workflow environments, and previously constructed an infrastructure that assumes conventional high-speed networking. Professor Kaiser is developing new algorithms and techniques to support mobile hosts. The research addresses prefetching based on semantic information on what the user is likely to do next drawn from the workflow definition tailoring the specific environment instance, to permit low-bandwidth and temporarily detached operation by participants in collaborative enterprises. Professor Duchamp is developing operating system support for intelligently prefetching files from a file system or objects from an objectbase. The goal is to improve performance and automate cache loading prior to disconnecting from the network, via two types of prefetching: The first is performed by the file system, transparent to applications; the file system ``learns'' file access patterns and uses these patterns to prefetch if/when the patterns recur. The second type of prefetching is application-directed; applications with knowledge of their future data accesses can use a special interface to instruct the operating system what to prefetch. Professor Kaiser's workflow system is the major example application for this type of prefetching. Both the collaborative work and operating systems directions require mobile notebook computers for development, testing, and measurement of the prototype support.
