Wide-area backbone networks of Internet Service Providers (ISPs) and cloud providers, such as Microsoft, are the workhorses of Internet traffic delivery. To support the explosive demand, efficient use of the physical layer is imperative. Bandwidth variable transceivers (BVTs) are an enabling technology for elastic optical networks. They can convert physical layer performance into network capacity using variable coded modulation, rate-adaptive FEC or parity coding, or other means. For such techniques to find application, however, we need to understand the potential gains in practical field deployments. Signals require sufficient performance above transmission engineering margins (additional penalties that take into account uncontrolled system time and configuration dependent phenomena) in order to move to higher capacity modulation. In this paper, we examine both the measured and simulated optical performance of Microsoft’s backbone network, and estimate the achievable benefits of BVTs.
Since the value of BVT is strongly dependent on the performance and available margins, we quantify the performance of a live backbone network and use that data to estimate the potential benefits. In particular, for a period of three months, we poll the signal Q-factor of all 100Gb/s PM-QPSK channels in the backbone network. From this, we can determine the performance of each channel relative to its minimum. We simulate the performance of 8-QAM and 16-QAM modulations in order to determine the propagation penalties associated with the fiber nonlinear response, which must be included due to its uncertainty. We then evaluate the number of paths that may be accessible using these formats and approximate further improvements from BVTs with a resolution of 25 Gb/s. Next, we consider the time and wavelength variations of Q-factor to determine whether fluctuations might be too rapid to adapt to as well as whether wavelength-dependent variation margins are accessible to BVT.