This paper analyzes three time to live (TTL)-based cache policies adopted in Web services. Specifically, authoritative, exclusive, and independent sources are defined. The authoritative source can provide a copy of the requested object with zero age. That is, objects are always fresh in the authoritative source. An exclusive source periodically updates objects from an authoritative source. Depending on the time of the request to the exclusive source, the age of the object is determined. Finally, an independent source forwards a request to high-level caches upon each miss. In this scenario, the age of the object is independently drawn from a uniform distribution. The notions of extended lifetime and rejuvenating, respectively representing delayed and expedited object refreshing, are introduced next. The paper continues to compare the performance of these three policies and their variations.
Cohen, Halperin, and Kaplan are experts in this area, and have published several related papers. This paper is no disappointment. It has an excellent introduction section. The objectives of the paper are explained clearly early on, and are reiterated as the reader proceeds. The paper contains quite elaborate mathematical treatments. The proofs presented are mostly solid and straightforward. Some equations are given without explanation, breaking the paper’s rhythm in places. However, this is only a minor issue. Interested readers should have no problem figuring them out. Overall, I think the paper makes a good contribution in laying out rules for configuring Web caches. In particular, some rules are not exactly intuitive, and this makes their discovery more valuable. I recommend this paper to those who wish to gain insight into this topic. Those who cannot spare the time to read through the entire paper can probably still learn quite a bit by reading the introduction! and conclusion sections.