Can observed entropy detect congestion in Ad-Hoc networks?
Xiaojie Liu, Ulrich Speidel · 2017
Congestion presents a significant challenge in ad hoc networks due to their unstructured and distributed nature. In most congestion detection schemes for such networks, the affected node itself detects whether it is congested or not. The detection approach proposed in this paper performs detection with information estimators from neighbouring uncongested nodes that may be able to relieve the congestion. The type of information estimator studied here is the entropy of the difference between subsequent sequence numbers of TCP packets sent to or via the node under congestion monitoring. Using the hidden node problem as background, this paper argues that entropy estimates are as sensitive to congestion as the observed retransmission rate, but easier to compute.