Admission control in data transfers over lightpaths

The availability of optical network infrastructure and appropriate user control software has recently made it possible for scientists to establish end-to-end circuits across multiple management domains in support of large data transfers. These high-performance data paths are typically provisioned over 10 Gigabit optical links, and accessed using ethernet encapsulation at Gigabit and 10 Gigabit rates. The resulting mixture of circuit sizes gives rise to resource conflicts whereby requests to allocate bandwidth partitions are blocked despite vast underutilization of the optical link. In an attempt to remedy this problem, we investigate intelligent admission control policies that consider the long-term effects of admission decisions. Using analytic techniques we show that the greedy policy, which accepts requests to allocate bandwidth partitions whenever sufficient bandwidth exists, is suboptimal in a pertinent scenario. We then consider dynamic online computation of the optimal admission control policy and show that the acceptance ratio of requests to establish end-to-end circuits can be improved by up to 19% on a fifteen-node network where the behaviour of each link is governed by a local optimization effort.