Abstract

This thesis proposes ideas for designing file system software for large, high-performance file server hardware we feel will be common in the middle to late nineties. In particular, the thesis examines the value and pragmatics of file systems with multiple file implementations. A file implementation determines how a file is represented in secondary storage and the procedures by which that representation is interpreted. A file system with multiple file implementations can use different implementations for different files. The thesis also proposes an allocation algorithm designed for a system with device parallelism and multiple file implementations, and it reports the results of a trace-driven simulation study evaluating the algorithm. The thesis proposes parameterizing file behavior to give users control over which implementation is used for a file without exposing the low-level details of implementations.

Citations