Caching dynamic contents with varying popularity

We study content caching in a cellular network consisting of a base station with a cache. New contents arrive in the network according to a Poisson process and the contents stay for exponentially distributed times. At any given time, all the contents in the network have the same popularity or request rate. Also, contents’ request rates are time-varying, instantaneous request rates being a decreasing function of the number of contents in the network. Precaching contents at the base station incurs a cost. However, fetching contents from the server on being requested incurs an even higher cost. We formulate caching problem as a Markov decision process and derive the optimal caching policy. We also propose a Reinforcement Learning based algorithm that yields precaching decisions when system parameters are unknown. Numerical results show that the proposed algorithm’s time averaged cost is close to the optimal average cost.