A POMDP-based caching algorithm for RAM of steaming media servers

Streaming media servers need to acquire data from local disk frequently. Because of the much lower reading/writing speed of disk than random access memory (RAM), good RAM caching policy can efficiently reduce disk I/O. The traditional technique is using a greedy and suboptimal algorithm based on the 0-1 knapsack problem model. In this paper we propose a partially observable Markov decision process (POMDP) model for the caching problem. Based on the model, the method of event-based optimization is applied in search of optimal randomized policy. The disk I/O, as the long-run average performance measure of a Markov chain, is optimized via applying the policy-gradient algorithm. The simulation results illustrate that the requiring data could be predicted better and lower disk I/O could be achieved by using the proposed method.