Performance Aspects of Distributed Caches Using TTL-Based Consistency

Web objects are stored and can be requested from numerous servers, including authoritative “origin” servers and caches. Objects can be modified only by their origin servers and weak consistency with cached copies is maintained by limiting their lifetime durations. Copies fetched from origin servers are received with maximum time-to-live (TTL) that equals their lifetime duration whereas copies obtained through a cache have shorter TTLs since their age (elapsed time since fetched from the origin) is deducted from their lifetime duration. A request served by a cache constitutes a hit if the cache has a fresh copy of the object. Otherwise, the request is considered a miss and is propagated to another server. Performance is measured by the number of requests constituting cache misses. It is evident that the number of cache misses depends on the age of the copies the cache receives. Thus, a cache that sends requests to another cache would suffer more misses than a cache that sends requests directly to an authoritative server. More subtly, the number of misses depends on the particular configuration of higher-level caches, e.g., whether one or more higher-level caches are used. Guided by practices for Web caching, we model and compare different configurations. We also analyze the effect of pre-term refreshes at high-level caches and extended lifetimes at low-level caches and reveal patterns that may seem counter-intuitive at first. Even though TTL-based consistency is very widely used, our work seems to be the first to formally model and analyze it. Our analysis yields insights and guidelines for improving the performance of Web caches.