Autonomous Traffic Engineering using Self-Configuring Link Weights

Abstract

Network operators use traffic engineering to control the flow of traffic across their networks. Existing TE methods establish static topologies offline, either by setting link weights or by configuring paths a priori. These methods require manual configuration and may not be robust in the face of failures. Some methods also require knowledge about traffic demands and may not be able to handle traffic fluctuations. Even when changes in demand are expected, operators must manually tune network configurations to prepare for them. Because adjusting configurations is difficult to get right, we start from an extreme design point, asking instead whether it is possible to perform traffic engineering online without having to perform any a priori configuration. Our traffic engineering technique, SculpTE, adapts to changing traffic demands by automatically configuring link weights in a stable manner. SculpTE balances load across the network by continually adjusting link weights to expose lightly-loaded paths. We evaluate SculpTE using a simple analytical model and simulations on realistic ISP network topologies. Our results show that SculpTE achieves excellent load balancing, responsiveness, and stability compared to state-of-the-art TE schemes, without requiring network operators to perform any offline configuration.