Accumulation-based congestion control and its applications
Shivkumar Kalyanaraman, Yong Qing Xia · 2004
The Internet is all about providing services to its applications. These services are defined in terms of packet delivery reliability, timeliness and flow bandwidth, etc. For example, IP provides hop-by-hop best-effort packet delivery; TCP adds end-to-end reliability. However, TCP/IP does not provide any bandwidth predictability. In fact, the bandwidth under TCP's loss-based congestion control algorithm is inherently oscillating. This is undesirable for some emerging applications that require stable and sufficient end-to-end bandwidth. To address this problem, we propose a model to use accumulation , buffered packets of a flow inside the (congested) network routers, as a congestion measure on which a family of congestion control schemes can be derived. We call this model accumulation-based congestion control (ACC). A class of control algorithms that have the same equilibrium properties exist. Consequently, we have the freedom to choose a proportional control policy with desirable stable bandwidth dynamics. The ACC model serves as a reference for packet-switching network implementations. We show that TCP Vegas [19] is one example scheme that fits into the model. Unfortunately, Vegas suffers from round trip propagation delay estimation error and reverse path congestion. We therefore design a new Monaco scheme that solves these problems by employing an out-of-band, receiver-based accumulation estimator, with two-priority-FIFO-queue support from the bottlenecks. We use extensive ns-2 simulations to validate that the static and dynamic performance of Monaco matches the theoretic results. One key issue regarding the ACC model in general, the scalability of the bottleneck buffer requirement and a solution using an adaptive virtual delay (AVD) queueing mechanism, are evaluated. Our scheme is incrementally deployable in that the AVD routers can work with other non-AVD droptail routers. We extend the ACC model as a closed-loop, data-plane building block to achieve a range of better bandwidth services than what TCP/IP provides. They include a weighted rate service for bandwidth differentiation and an expected minimum rate service for bandwidth guarantee. Our approach is to map bandwidth allocation onto accumulation allocation and then apportion accumulation among the competing flows appropriately. Both services are meaningful in steady state and can be modelled as moving the equilibrium in Kelly's nonlinear optimization framework for network congestion control [65].