Short-Sighted Routing, or When Less is More

Due to the high costs of wide area links, traffic engineering is frequently used to optimize wide area networks' performance. In recent years, architectures based on software defined networking approaches are being proposed as better ways of implementing network control and management. However, network optimization requires timely acquisition of statistics on network traffic demands by SDN controllers, as well as timely reconfiguration of network paths. Both goals may only be achieved with a scalable control plane. In this article, we present an approach meant to trade complexity and scalability of the control plane for routing optimality. Our experiments suggest that the performance of provider networks is nearly optimal, even when load balancing parameters have been computed using a slightly outdated vision of traffic demands. We named this approach short-sighted routing (SSR). Our results also show that performance optimality in many provider networks that use load balancing over several paths is not affected in a significant way by relaxing the timeliness of network traffic demands acquisition and the corresponding load balancing reconfiguration. This has allowed us to conclude that dynamic centralized network optimization and management, in accordance with an SDN approach, is also quite realistic in service provider networks.