Imperfect Bandwidth-Sharing Policies using Network Calculus

Bandwidth-sharing policies aim at enforcing fairness among several classes of traffic by reserving one share of the available bandwidth for each class. They recently attracted a lot of attention, in particular in the context of time-sensitive networking. One advantage of these policies is their simplicity of implementation, especially the Round-Robin policies. Moreover, if a traffic class does not use all its reserved bandwidth, the other classes can use the unused part. Recent works greatly improve the delay bounds by considering this phenomenon for the Deficit Round-Robin policy, when the knowledge about the incoming traffic is available. There was also an attempt to use this approach for Weighted Round-Robin policy. Unfortunately, this was incorrect, mainly due to the variability of the packet lengths and the share of the bandwidth is not perfectly known. In this paper, we propose a generalization of bandwidth-sharing policies, that we call imperfect bandwidth sharing. We compute per-class service guarantees that correct and improves the state of the art on Round-Robin policies, and assess through numerical experiments the algorithmic and performance gain of our solution.