Intelligent caching for remote file service

Limitations of current disk block caching strategies are discussed. A model for providing remote file service using knowledge-based caching algorithms is proposed. The knowledge-based algorithms generate expectations of user process behavior which are used to provide hints to the file server. Te research involved gathering trace data from a modified Unix kernel and conducting trace-driven simulation of remote file server models. Performance improvements of up to 340% were observed for knowledge-based caching in simulated file service. Comparisons are made between conventional, knowledge-based, and optimal models. Extensions to general caching are discussed