A Study Of Congestion Control Mechanisms In Delay Tolerant Networks

In Delay Tolerant Networks (DTNs), nodes communicate with each other based uponintermittent or opportunistic connectivity. Because the propagation delay may be extremelylarge and end-to-end path cannot be assured, the data transmission mechanism in DTN isvery different with that in the Internet. Other than packet switching, DTN employs virtualmessage switching and transfers message bundles hop by hop. At the same time, to tacklewith the unreliability of underlying links, most routing protocols in DTN spread multiplereplicas to improve the probability of successful delivery.These peculiarities of DTN make the congestion issue even more prominent. The in-termittent connectivity causes messages to accumulate transiently in the node buffer andcauses fast depletion of storage resource. What’s more, excessive replication brings toomuch network overhead. Even if a message has already been delivered, its replicas maystill be buffered in intermediate nodes. Congestion causes packet loss and rise of queuingdelay, degrading the overall delivery probability. It also wastes a lot of storage, energy andbandwidth resources.Traditional congestion control mechanisms based on stop and wait, sliding window inthe Internet cannot be applied directly to DTN. Besides, they overlook multiple replications,delivery probability, which is specific to DTN. Although a few works have discussed thebuffer management problem in DTN, they mostly are based upon single metrics and specificnetwork model, lacking necessary theoretic ground and realistic operability. We analysis theincrement of delivery probability caused by message forwarding and discarding and presenta new buffer management scheme based on probability increment (PI), which interpret theinfluence of message replicas and its TTL value. A detailed analysis is performed on theexponential inter-contact time model. A message status management module is also devisedto record and approximate global knowledge. For probabilistic routing, the deviation ofdelivery predictability may be enlarged which makes PI unsuitable. We analyze theimportance of contacts and approximate message replicas with local knowledge and present a new buffer management mechanism for probabilistic routing. Extensive simulations showthat the proposed buffer management algorithm can improve network performance greatly.