Traffic management and QoS adaptation in heterogeneous data networks

The future network infrastructure will be very heterogeneous. It will consist of end computing systems with widely varying capabilities, interconnected by networks which offer very different services. Since the key to the performance of the network is control, both support at the network level (integrated services networks) and a QoS (Quality of Service) management environment are required in order to deliver data with a guaranteed level of service. It is also crucial to allow applications to monitor the underlying system's QoS and adapt to the type of large fluctuations in QoS which are a characteristic of mobile environments.; This dissertation investigates the importance and requirements to provide QoS in heterogeneous data networks. In addition, two classes of flow control mechanisms, the controllable network and end-to-end flow control, with opposite design philosophies are compared. The flow control schemes utilized in controllable networks are designed with the explicit goal to make the network and transport layers to be fully controllable and observable while end-to-end flow control is not. The simulations using OPNET show that rate based flow control in controllable networks is superior over end-to-end congestion control in terms of efficiency, fairness and convergence. Last, the primal contribution of this dissertation is to provide a framework for traffic management and QoS adaptation for heterogeneous data networks. The proposed scheme, MAQ (Multiplexed Adaptive Queuing), is based on proper feedback with explicit goal to make the network observable and controllable. The scheme is a scalable closed-loop control protocol based on dynamic priority labeling of packets at edge nodes and bandwidth assignment at both edge and core nodes using the rate-based feedback from the network. Various connections with a wide spectrum of end-to-end QoS requirements are statistically multiplexed to the extent that their QoS requirements can be satisfied while still maintaining simplicity and scalability of scheduling and buffer management. Individual flows are mapped to fine-grain QoS classes at the edge nodes, while at the core nodes, coarse-grain QoS is provided by aggregating flows into limited number of priority queues. Therefore, the resolution of relative priorities among different traffic streams is pushed to the edge, and the design of high-speed switching fabric in the core is simplified. Analysis and simulation studies demonstrate the effectiveness of the proposed scheme.