Maximizing Hit Ratio in Finite-Buffer Caching with Request Delay Information: An MDP Approach

Abstract

Caching is a promising technology that holds the potential of substantially improving the bandwidth efficiency and reducing peak data traffic. As a result, the maximization of cache hit ratio in proactive caching has attracted much attention most recently. Unfortunately, the hit ratio can be very limited due to the constrained buffer size. In this paper, we are interested in maximizing the hit ratio for users with limited buffer size by exploiting the prediction of a user's request time for content files, also referred to as the request delay information (RDI). More specifically, the hit ratio is maximized by keeping the popular and storage-efficient content files in a receiver buffer. To achieve this goal, we formulate an infinite horizon Markov decision problem (MDP) that can be efficiently solved by a value iteration algorithm. For more practical applications, we present a heuristic caching policy that can greatly reduce the computational complexity when the buffer size is large and the content arrival rate is high, thereby holding a great potential in practice. Simulation results show that the RDI may bring significant hit ratio gain, especially in small buffer scenarios.