Temporal locality and its impact on Web proxy cache performance

This paper studies temporal locality characteristics present in document referencing behavior at Web proxies, and the impact of this temporal locality on document caching. First, drift measures are developed to characterize how the popularity profile of “hot” documents changes on a day-to-day basis. Experiments show that although there is considerable “hot set” drift, there is also a significant number of documents that have long-term popularity. Second, a measure of short-term temporal locality is developed that characterizes the relationship between recent-past and near-future document references. Using this measure, it is established that temporal locality arising out of the correlations between document references in the recent-past and near-future does exist for popular documents. r objective of this paper is to determine whether or not Web document references at proxy caches can be modeled as independent and identically distributed random events. Trace-driven simulations using empirical and synthetic traces (with varying degrees of temporal locality) show that temporal locality is an important factor in cache performance. The caching simulation results also show that temporal locality arising out of short-term correlations between references is important only for small caches. For large caches, a synthetic workload generated by applying the Independent Reference Model on a day-to-day basis gives performance very similar to that obtained for empirical traces.