The Pair Cache Problem

Abstract

This thesis investigates the pair cache problem, a unique variation of the classic cache replacement problem where each element is stored in two pages, and the cache only needs one of these pages to respond to a query. The thesis formalizes the definition of the pair cache problem, explores relevant historical background, and investigates both offline and online cases. In the offline case, we show that approximating the problem with a factor less than two is NP-hard, while in the online case, we extend the FIFO algorithm and prove that it works as a 4-factor approximation. The proposed pair cache scheme has practical applications in systems where data retrieval times are slow, and most processes require information stored in two different data sources. The goal of a pair cache management algorithm is to minimize the number of retrievals by deciding which page to retrieve on a miss and which page to discard when the cache is full.