A fuzzy block replacement algorithm for disk caches

Abstract

In this paper, we present a new block replacement algorithm for disk caches. Our Fuzzy Block Replacement Algorithm (fuzzy block-RA) utilizes nine fuzzy rules to pick out the block to be evicted when an I/O reference causes a miss on a filled up disk cache. The decision making process is based on the age of a block and the frequency at which references are made to it. Optimization of the knowledge base to specific applications is easily possible by modifying the inference rules or tuning of the membership functions. The simulations performed using our algorithm shows that the miss rate improves with a fuzzy block-RA when compared with standard algorithms. The use of disk traces gathered from a commercial disk subsystem in the simulation ensures the practical validity of the results.