Multipoint-to-point communication allows a group of sources to transfer data to one destination. A major requirement of flow control for such connections is to ensure a fair allocation of resources while maintaining a high level of resource utilization. This work treats multipoint-to-point flow control as a multiple-objective optimization problem and presents a theoretical centralized model as well as a distributed algorithm to compute rate allocations based on this global optimization. Three control objectives have been identified as critical to the flow control of multipoint-to-point connections: overall network throughput, fairness amongst sources, and fairness amongst groups.; The theoretical model is a linearly constrained quadratic programming model with an objective of minimizing weighted sums of individual objective functions. The weighting factors become tuning factors with which decision makers can set their decision preferences. It was shown that the three objectives may indeed be conflicting with each other and, by varying the values of tuning factors, an optimum rate allocation can be achieved to realize many flavors of objective mix.; The distributed algorithm attempts to implement the centralized model in a distributed environment. A resource pricing with an aggregate utility maximization scheme was used to allocate bandwidth to maximize overall throughput. The algorithm is integrated with an explicit rate indication algorithm that optimizes resource allocation based on the source fairness criteria. It was shown that this algorithm attains similar result to the theoretical model and is also tunable.; This thesis shows that multiple-objective optimization-based rate allocation is feasible, flexible, and powerful, especially in situations where we are able to trade some objectives for others.
展开▼
