Using performance maps to understand the behavior of Web caching policies

Abstract

One of the most important differences between traditional caching and Web caching is the high variability of the size of Web objects. As a consequence of that variability, we use two metrics to assess performance of Web caching replacement policies: hit ratio (HR) and byte hit ratio (BHR). The paper presents an analytical model relating these two performance metrics. We show that the relation between hit ratio and byte hit ratio is a function of the mean size of the Web cache server responses and of the mean size of the hit responses. Based on the analytical relationship, we introduce the concept of performance map for a given workload. Performance maps are useful tools for designing and comparing the performance of different replacement policies that deal with objects of variable size, such as the Web caches. Performance maps are applied to some actual workloads that have been used as de-facto standards for comparing the performance of Web caching management algorithms.