Optimal Buffer Management Policies in DTNs: A POMDP Approach

Using Delay Tolerant Networks (DTNs) would facilitate the connection of devices and areas across the world that are under-served by current networks. The DTNs are based on the concept of store-carry-and-forward protocols. A node may store a message in its buffer and carry it for a short or a long period of time, until an appropriate forwarding opportunity arises. A critical challenge is to determine routes through the network without even having an end-to-end connection. In order to increase the probability of message delivery, a known approach is implemented using epidemic message replication. This combination of long-term storage and message replication imposes a high storage and bandwidth overhead. Thus, efficient scheduling and dropping policies are necessary to decide which messages should be discarded when nodes' buffers operate close to their capacity. If a relay buffer is full and needs to store a new packet, it has to decide either to keep the current message or to drop it. This decision depends on the number of transmissions of the message in its buffer and the message which has just arrived. In this paper, a Partially Observed Markov Decision Process (POMDP) framework is proposed to solve the problem of buffer management in DTNs. This modeling technique predicts some properties of the optimal buffer management policy. In addition, numerical examples are presented to illustrate the findings.