The authors address the problem of flow control and routing of real-time traffic in a network, where messages must arrive at their destination within given deadlines if they are not to be considered lost. Performance in this case is measured in terms of the probability of losing a message. For the case of n parallel links, the problem is formulated as one of optimal flow allocation and solved under general conditions. It is shown that for a FCFS (first-come first-served) service discipline an admission policy rejecting messages before link assignment is optimal when the load exceeds a critical value. Thus, the authors take advantage of the fact that if some messages will exceed their deadlines anyway, it is beneficial not to admit them in the first place. An efficient algorithm for explicitly solving the problem is presented and specific examples are analyzed. The authors also discuss the applicability of online algorithms for this problem when modeling assumptions cannot be made.
展开▼
