Bandwidth calendaring: Dynamic services scheduling over Software Defined Networks

Software Defined Networking enables centralized network control and hence paves the way for new services that use network resources more efficiently. Bandwidth Calendaring (BWC) is a typical such example that exploits the knowledge of future to optimally pack the arising demands over the network. In this paper, we consider a generic BWC instance, where a carrier network operator has to accommodate at minimum cost demands of predetermined, but time-varying, bandwidth requirements. Some of the demands may be flexible, i.e., can be scheduled within a specific time window. We demonstrate that the resulting problem is NP-hard and we propose a scalable problem decomposition based on column generation. Our numerical results reveal that the proposed solution approach is near-optimal and outperforms state-of-the art methods based on relaxation and randomized rounding by more than 20% in terms of network cost.