Lighting fibers in a dark network

摘要

We consider the problem of network design in transparent, or clear channel, optical networks associated with wavelength-division multiplexing (WDM). We focus on the class of traffic engineering models known as routing, wavelength, and capacity assignment problems. Here, in contrast to traditional networks, traffic flow paths must also be assigned an end-to-end wavelength. This additional requirement means that there can be an increased cost associated with optimal capacity allocations for such WDM-flows. In general, this can be arbitrarily worse than traditional network designs. We argue that in order to evaluate the benefit of different switch technologies, a good benchmark is to measure the increase in costs purely in terms of link capacity, we call this the cost of transparency. Experimental research shows that this cost is small in multifiber networks with modest switching functionality at the nodes. We present theoretical justification for why this occurs, and prove that in multiwavelength multifiber transparent networks the cost of transparency all but disappears if there is moderate traffic load. Our arguments are based on efficient heuristics that may also be useful for more complex network optimizations. This suggests that the cost savings from using wavelength converters is significant only in young networks with relatively few fibers lit. Such savings may, thus, be small relative to the initial capital expense involved in installing wavelength conversion.