The paper is devoted to caching of popular
multimedia and Web
contents in Internet. We study the Cluster Caching Rule (CCR) recently proposed by the authors. It is based on
the idea to store only popular contents arising in clusters of related popularity processes. Such clusters defined as consecutive exceedances of popularity indices over a high threshold are caused by dependence in the inter-request times of the objects
and, hence, their related popularity processes. We compare the CCR with the well-known Time-To-Live (TTL) and Least-Recently-Used (LRU) caching schemes. We model the request process for objects as a mixture of Poisson and Markov processes with a heavy-tailed noise. We focus on the hit probability as a main characteristic of a caching rule and introduce cache effectiveness as a new metric. Then the dependence of the hit probability on the cache size is studied by simulation.
