Simulation evaluation of a relative frequency metric for web cache replacement policies

Abstract

In this paper, we interest in web cache replacement policies namely “frequency-aware” policies that provide, generally, the best results in term of data movement reduction in the network. For the simple reason that they take into account one of the most significant web traffic characteristic “the access frequency”. However, the access frequency suffers from two main problems namely one-timer documents existence and cache pollution. Therefore, our aim is to replace the traditional frequency with a relative frequency; calculated using the access number and the document lifetime in the cache. Although the idea already exists in the literature, we strive to validate the relative frequency efficiency for the web proxy replacement policies. In this work, we implement three replacement policies namely least frequently used (LFU), least frequently used with dynamic aging (LFU-DA) and Greedy dual size frequency (GDSF). As well, their versions enhanced with relative frequency namely LFRU, LFRU-DA and GDSFR respectively are implemented and evaluated using synthetic and real workload. The simulation results show that the relative frequency is more effective, in terms of hit rate and byte hit rate, than the access number; i.e., the traditional frequency. Moreover, the simulation proves that the relative frequency solves the access frequency problems.