Modeling the dynamics of caching in content-based publish/subscribe systems

This paper considers cache dimensioning in the context of publish/subscribe (pub/sub) systems. We assume that each broker is equipped with a limited capacity cache and it decides upon a policy for caching and prioritizing messages. By using a request mechanism defined on top of the native pub/sub communication, a client may also request earlier published information. To study the survival time of published messages, a Markovian system model capturing the essential dynamics is defined. The model has a modular generic form which admits a variety of different policies and thus enables the calculation of their performance. For systems without message replication between the caching brokers, the distribution of message survival time is found using matrix analytic methods for solving absorbing Markov chains. For the general problem with messages copied from caches, we propose a heuristic approximation based on estimating the mean rate of copies. The approximate model is evaluated by a discrete event simulator and it is shown that for a wide set of parameters, the approximation provides a good basis for dimensioning the caches in the content-based pub/sub systems.