Optimal Caching Policy of Stochastic Updating Information in Delay Tolerant Networks

Abstract

To increase the speed of information retrieval, one message may have multiple replicas in Delay Tolerant Networks (DTN). In this paper, we adopt a discrete time model and focus on the caching policy of stochastic updating information. In particular, the source creates new version in every time slot with certain probability. New version is usually more useful than the older one. We use a utility function to denote the availability of different versions. To constrain the number of replicas, we propose a probabilistic management policy and nodes to discard information with certain probability determined by the version of the information. Our objective is to find the best value of the probability to maximize the total utility value. Because new version is created with certain probability, nodes other than the source may not know whether the information stored in them is the latest version. Therefore, they can make decisions only according to the local state and decisions based on the local state can be seen as local-policy. We also explore the global-policy, that is, nodes understand the real state. We prove that the optimal policies in both cases conform to the threshold form. Simulations based on both synthetic and real motion traces show the accuracy of our theoretical model. Surprisingly, numerical results show that local-policy is better than the global-policy in some cases.