MS Project

Implementing Prefetch Algorithm in Hint-based Collaborative Caching

Masters Project

Department of Computer Science

Rochester Institute of Technology

Abstract

The increasing size of data and growing demand for its faster access rate has pushed the boundaries of I/O. This gave birth to collaborative caching which exploits the potential of faster network transfer rate and speedy computing processors. Among the various types of proposed algorithms, hint-based collaborative caching by Sarkar and Hartman is the most. efficient [7]. Although its architecture and design appear to be flawless it could still be enhanced with introduction of prefetching scheme.

This project integrates two prefetching schemes namely LRU-Lookahead and PrefixSpan into hint-based collaborative caching algorithm. These algorithms have been used in area of parallel disk access system and RAM grid co-operative cache respectively [4, 3]. We have tweaked the hint-based algorithm to accommodate these prefetching schemes. The analysis of the results over a simulated environment show promising results when compared with original algorithm. In our experiments we have considered a block access patterns, where blocks can be read sequentially or non sequentially. Introduction of LRU-Lookahead proves to be beneficent in sequential access pattern whereas PrefixSpan performs better under non sequential access pattern conditions.