In the proceedings of the 13th International Conference on the Design of Reliable Communication Networks (DRCN 2017), pp. 39-46, Mar. 2017.Winner of the best paper award.
Abstract The possibility of flexibly assigning spectrum resources with channels of different sizes greatly improves the spectral efficiency of optical networks. The drawback of this flexibility is the risk of spectrum fragmentation. We study this problem in the two-service scenario. Our first contribution consists of exact Markov models for different assignment policies. Since these exact models do not scale to large systems, we then extend an approximate, reduced-state model that is available in the literature. In addition, we introduce a Markov model that uses imprecise probabilities, which allows us to derive upper and lower bounds on blocking probabilities without needing to specify an assignment policy. The obtained imprecise Markov chain can be used to evaluate the precision of approximate reduced-state models as well as to provide policy-free performance bounds.