Is it worth to keep an all optical OBS data plane?

Optical burst switching (OBS) has been proposed a decade ago to enhance the resource utilization of the current OCS technology throughout statistical multiplexing. Therein the lack of optical memory is sidestepped by advanced signaling protocols that cannot, however, prevent contentions. The contention probability is highly related to the traffic profile in the core network for which the Engset model has been shown relevant. In OBS, as the data plane is transparent, the traffic profile in the core network is completely defined by the emission process. In this paper, we describe the emission process and identify configurations that take advantage of the Engset model. We then propose a hybrid architecture that features burst re-aggregation in some intermediate nodes to exploit the favorable emission configurations. This variation of OBS requires only a slight modification of the core nodes and significantly impacts the performances: Simulations performed on the EONET network show a significant loss rate reduction, whereas the delay is in the worst case slightly increased.