Locality of Reference in an Hierarchy of Web Caches

Abstract

This work presents an extensive evaluation of the request filtering in hierarchy of proxy caches. Using the recently proposed ADF (Aggregation, Disaggregation and Filtering) model as well as entropy as metric for Web traffic characterization, we evaluate how locality of reference changes as the streams of requests pass through a hierarchy of caches. Moreover, we propose the use of average entropy for comparing the locality of reference of different streams and present how a proxy server can dynamically calculate the entropy of its incoming request stream.

On workload merging and filtering effects in hierarchical wireless media streaming

“…Previous research on Web workloads shows that reference locality can be exploited in Web cache design. Reference locality has two aspects: the popularity of requested objects, and temporal correlations in the workload [2,7,11,13,16]. As proposed in [11], we use entropy as one of the metrics for popularity.…”

“…Entropy H(X) was proposed in [7,11] as a way to measure the uniformity or non-uniformity of the popularity distribution in a workload. More specifically, entropy captures the skew of requests to a set of objects.…”