Cost-effective design of wdm optical networks for static and dynamic traffic

Current optical networks use fiber-optic to replace copper wiring and have achieved a speed of 10 to 100 times faster than their electronic counterparts. However, they have only realized a fraction of potential bandwidth of fiber optics because the data have to go through electronic switches, and therefore limited by electronic speed. In this research, we study wavelength division multiplexed (WDM) all-optical networks, which, by eliminating electronic bottlenecks, may improve the capacity of current optical networks by a factor of 1,000. In WDM networks, a physical link is multiplexed using multiple wavelengths. We propose a framework for the cost-effective design of WDM networks and evaluate two representative approaches, namely Link Multiplexing (LM), in which different wavelengths could be used on different links along the path of a connection, and Path Multiplexing (PM), in which the same wavelength has to be used along the path. We first analyze the effect of the two approaches on some near term applications. Particularly, we study the off-line and on-line permutation embedding and scheduling capabilities of WDM networks using the two approaches. We find that PM is almost as effective as LM for off-line permutation embedding and scheduling while the advantage of LM may become more obvious for on-line cases. We then extend the study by considering future applications requiring frequent on-demand network reconfigurations. We focus on protocols for establishing all-optical paths under distributed control because of its scalability and reliability. We start with a basic reservation scheme and explore various dimensions to improve performance, including source initiated reservation (SIR) vs. destination initiated reservation (DIR), one-way reservation vs. two-way reservation, simple control vs. informed control, spendthrift reservation vs. frugal reservation. Our simulation results show that DIR outperforms SIR significantly while its complexity is comparable with SIR. One-way reservation, informed control, frugal reservation can improve performance further at the cost of extra hardware and more complex software. In addition, we find that the advantage of LM over PM is more obvious under distributed control, especially for networks with large propagation delay, where LM could outperform PM by 100% in terms of maximum link utilization achieved.