Speaker: Nizar M (EE11S040)
We study the Rate Control Protocol (RCP), which uses rate-based feedback to control flows and network congestion. RCP assigns high and fair starting rates to flows, which is estimated using two forms of feedback from queues in routers: rate mismatch and queue size.
We first study some equilibrium properties. A key design question is whether we need both forms of feedback. In particular, if the feedback based on queue size is helpful or not. Using simulations, we highlight that for small bandwidth-delay product environments, the presence of queue feedback can lead to the emergence of limit cycles in the queue size. We consider a non-linear model of RCP which is coupled with a fluid model of the queue size. For this model, we conduct a local stability and local bifurcation analysis. We highlight that the model undergoes a Hopf type bifurcation. Further using Poincare normal forms and center manifold theorem, we explicitly characterize the type of Hopf bifurcation. We show that the system, in the presence of queue feedback, would undergo a subcritical Hopf bifurcation for some values of the RCP parameters. We also outline some design considerations to ensure the stability of the queue size.
We further analyse some dynamic properties of the RCP system. In particular, the management of flows as they arrive into a RCP network. To that end we investigate a recently proposed admission management process. We conduct hardware based experiments using a modified NetFPGA RCP implementation of the process. Our study shows that the proposed process appears to be attractive for dealing with the admission of new flows, while maintaining small queues. Additionally, the algorithm allows routers to maintain a specified target link utilization. Such stable and small queues would help reduce latency. Further, by employing the admission management algorithm, buffers in routers may be dimensioned to be much smaller than the current, bandwidth-delay product, buffer sizing rule.