An effective buffer management in delay tolerant networks based on a partially observed Markov decision process framework

The delay tolerant networks work in communication environments that are a subject to delays and distributions. Mobile nodes in DTN can communicate with each other even though there is no permanent link between the source and the destination. To have a better rate of delivery of messages in DTN, the long-term storage and message replication require an efficient buffer management. In this paper, we propose a partially observed Markov decision process framework to solve the problem of buffer management. The node uses messages' transmissions number to decide which message to drop. We look for an efficient strategy that a node can follow to maximise its average reward, while taking into account the energy penalties in different levels of cooperation. For this, we have analytically determined the optimal policy's interval of our proposal. Furthermore, we use ONE simulator to compare our findings to traditional buffer management policies under different mobility models.