Energy Saving Through Traffic Profiling in Self-Optimizing Optical Networks

An increasing fraction of the electrical energy produced in western countries is being consumed by Internet infrastructure; reducing its energy footprint is therefore of utmost importance for the scalability of the Internet. We address optical transport backbones and propose a novel method to reduce the energy consumed by dynamically adjusting the number of active optical carriers to support the short-term load of the network with a small and controllable margin. This is achieved in a nondisruptive manner that does not interact with routing strategies and does not rely on any specific control plane, but exploits automated traffic profiling and prediction of the well-known circadian traffic cycle. The proposed approach works with both fixed and flexible grid optical networks. We describe a method to automatically learn these patterns and multiple techniques to predict incoming traffic. Furthermore, we present an algorithm that tunes the parameters of the proposed system in order to achieve a target a posteriori probability of causing traffic losses. The behavior of the system is studied, using simulations, under a variety of conditions. Results show that the proposed prediction algorithms can significantly reduce the number of active optical carriers, even in nonoptimal scenarios, while guaranteeing low traffic losses.