Impact of One-Timer/N-Timer Object Classification on the Performance of Web Cache Replacement Algorithms

Abstract

Web cache replacement Algorithms proposed in the literature try to maximize the Hit Ratio (HR), the Byte Hit Ratio (BHR), and the Delay Saving Ratio (DSR). However, even with an infinite Web cache storage capacity, values of these metrics could not exceed 70% most of the time. This is due to the fact that, given a workload, the first reference to an object is always a miss. Moreover, a statistical analysis of the workload shows that as much as 76% of objects are One-Timers (OT), i.e. they are referenced only once. Caching OT objects usually degrade the performance of all Web cache replacement algorithms. In fact, it may cause the eviction of N-Timer (NT) objects and hence increases the number of misses. In this paper, we present a technique to classify whether a cached object is an OT or not. We show through simulation that classification may significantly enhance the performance of replacement algorithms with respect to the HR, the BHR and the DSR.