Caching is applied to provide requested documents or contents quickly from the cache (a short memory).
We consider the Cluster Caching Rule policy proposed recently in Markovich (2015). The idea of the rule is to keep only highly popular contents in the cache. Due to dependency in the inter-request process and heavy-tailed distributed inter-arrival and inter-request times, such frequently requested documents arise in clusters of popularity. The corresponding clusters of documents are loaded in the cache. If the requested document is present in the previous cluster, then it stays further in the cache. Otherwise, it is evicted from the cache. We propose the hit/miss probabilities of such caching policy.
The cache size estimation is considered. A mixture of $m-$dependent Markov and Poisson renewal processes is proposed as an example of an inter-request times model.