Improving RWA-OBS formulation and solution
Résumé
Loss-less transfers are commonly associated with circuit switching: circuit switching in all-optical networks (OCS) prevent from collisions by avoiding multiplexing below the wave-length granularity. Spectral multiplexing can be combined with time multiplexing in opaque networks at the expense of additional equipment (MSPP) and complex synchronization issues. Optical burst switching has been proposed a decade ago as a promising alternative to OCS: Its asynchronous nature and its arbitrary granularity open the possibility of statistical multiplexing, but it is not well suited to handle loss-less requirements. Wavelength routing and TDM can still be performed at the expense of a slight modification of the control plane, but those solutions override the OBS flexibility and avoid the statistical multiplexing. The RWA-OBS problem tends to solve the drawbacks of both WR-OBS and SOBS. Based on several properties related to the in advance reservation and the data plane transparency of OBS, it integrates particular cases where two flows can be multiplexed on a given wavelength in a genuine OBS fashion without loss. We first enhance the original RWA-OBS formulation and then describe an iterative greedy heuristic and a column generation approach to improve the computing time and the scalability of the model. Experiments are conducted on two topologies to illustrate the potential of RWA-OBS that can compete with SOBS without requiring synchronization. The new proposed methods are then compared in terms of computing times and qualities. The efficiency of the column generation approach allows the integration of an additional constraint to control the compromise between the burst insertion delay and the throughput.