We present a new model that describes the queueing process of a communication network when data sources use window flow control. The model takes into account burstiness in sub-RTT timescales and instantaneous rate difference of a flow at different links. It is generic and independent of actual source flow control algorithms. We then apply the model to study dynamical behaviors of delay based congestion control protocols. We show that the ratios of RTTs are critical to the stability of such systems and discover previous unknown modes of instability. For a general network with multiple links, we demonstrate that due to the spatial propagation of oscillations, instability is possible even when all flows have the same RTTs. Closed loop packet-level simulations and testbed measurements are provided to verify the model and all of the predictions. At the end, we will discuss the challenges and opportunities that this model brings, in particular, the need of proper macroscopic parameters and possible hybrid simulators that work on more than one timescales.
Copyright. All Rights Reserved.