Congestion modeling in graph-routed Delay Tolerant Networks with Predictive Capacity Consumption

摘要

We present Predictive Capacity Consumption (PCC), a congestion modeling extension to graph-based routing protocols. This extension provides a solution to the problem of flow control in Delay-Tolerant Networks (DTNs) and other overlays that can neither synchronize physical link state across the network nor negotiate bandwidth consumption bridging heterogeneous link layers. PCC enables the construction of a distributed, predictive congestion model independent of the underlying link layer without requiring excessive broadcasts or other mechanisms unfeasible in DTNs. PCC examines information generated by routing protocols and adjusts local routing graphs to account for predicted message paths, correcting for downstream congestion and message retransmission. Unlike other mechanisms, the flow control provided by PCC can be implemented anywhere a graph-based routing methodology is used and the adoption of this method requires only minor modification to the in-situ routing framework. We describe the PCC algorithm, analyze its operation, and demonstrate its performance by simulating multiple data streams driving a set of constrained networks to saturation. The simulation results show that PCC improves the throughput of the network by 97% over table routing approaches and by 37% over graph routing approaches without congestion models.