A performance analysis of a cache-based file prediction protocol for mobile file systems

Abstract

File prediction is the process of bringing necessary files into cache before they are needed, while file caching is the process of maintaining required data in memory for future reuse. Our combined concept, which is the cache-based file prediction, is designed to support the main objectives of the Paradise File System, which are to provide high available and reliable storage for files, and guarantees that file operations are executed in spite of concurrency and failures. In this paper, we show that our cachebased file prediction scheme is efficient and practical by describing its design and implementation. Moreover, we report on its performance evaluation using a cluster of workstations. Our results indicate clearly that our design exhibits a significant degree of files availability and prediction accuracy.