Caching Techniques for Parallel I/O Servicing

Parallel and distributed systems archi tectures support parallel I O components Caching has been applied to distributed I O subsystems as a standard solution to the problems of fastening data accessibility and increasing data reliability Cache consistency mechanisms have been implemented in order to in uence the cache usefulness in a posi tive way This paper presents a new caching tech nique based on the genetic algorithm idea and exam ines the e ect of this technique on the parallel I O cache consistency and updating process Cached data blocks on parallel disks are considered as a population evolving over simulated time and are updated at regular intervals towards an improved cache content The proposed cache update scheme is compared with the LRU caching scheme which has been widely adopted The proposed technique shows improved performance compared to conven tional caching under simulation runs for various workloads