Traffic grooming in OBS networks based on virtual optical memories

Abstract

In this paper, we investigate the effect of grooming segment-based traffic on statistical multiplexing gain using simplified edge-node models of Optical Burst Switching (OBS) networks. The objective is to multiplex different bursts on the same output wavelength in order to optimize the bandwidth utilization on the optical fiber. We introduce a particular structure of the Burst Header Packet (BHP) in order to support a per segment processing at the core nodes. An algorithm for building a new BHP for the output multiplexed traffic is developed. A subsequent dropping strategy based on the availability of Virtual Optical Memory (VOM©) resources and the maximum delay a segment can have is proposed. Unlike the approaches that have been proposed in the literature for traffic multiplexing in OBS networks, the proposed grooming strategy is executed at the core nodes (instead of edge nodes) resulting in a more efficient utilization of the network resources. The effect of traffic variability and VOM© size on statistical multiplexing gain and bandwidth savings are evaluated. The evolution of the bandwidth utilization, loss probability, and transmission delay according to the VOM© characteristics is also studied.