Robust Energy-Aware Routing with Uncertain Traffic Demands

Energy conservation has become a major challenge to the Internet. Switching part of components into sleep mode is an effective way for energy conservation. Many existing approaches compute routing based on traffic matrices, to balance energy saving and traffic engineering goals, e.g., the maximum link utilization ratio (MLUR). However, accurate traffic matrices are difficult to obtain and change frequently, resulting in lack of routing stability and robustness. We propose to find one energy-aware routing robust to a set of traffic matrices. Such demand-oblivious routing problem has been studied without energy conservation, but it becomes more challenging when energy is considered. To overcome the challenges, we define a metric that reflects the MLUR distance from a routing to the optimal routing under certain energy conservation requirement. We model the problem of minimizing the metric, and analyze the upper bounds. Then, we propose a robust energy-aware routing (REAR) scheme to solve the problem by selecting sleeping links and computing the routing, based on a classical demand-oblivious routing algorithm. We evaluate our algorithms by simulations on real topologies. The results show that REAR is much more robust than existing approaches while more than 35\% line card power can be saved.