Towards Resource-Optimal Routing Plans for Real-Time Traffic

We discuss the issue of computing resource-optimal routing plans in a network domain. Given a number of known traffic demands, with associated required delays, we discuss how to route them and allocate resources for them at each node so that the demands are satisfied. While a globally optimal routing plan requires joint computation of the paths and of the associated resources (which was claimed to be NP-hard), in this paper we stick to existing approaches for path computation, and use mathematical programming to model resource allocation once the paths are computed. We show that the problem is either convex or non-convex, depending on the scheduling algorithms adopted at the nodes. Our results show that, by computing resources per-path, instead of globally, the available capacity can be exceeded even at surprisingly low utilizations.