Dynamic Traffic Engineering in SDN: A Topology-Aware Approach to Network Management, Congestion Avoidance, Latency Optimization, and Failure Resilience

Abstract

The Software Defined Network (SDN) concept separates the control plane from the data plane giving a global view of the whole network. This gives us the opportunity to perform traffic engineering that considers at once all these parameters: the traffic engineering that is based not only on congestion optimization, but also on latency minimization and reliability of communication. In this paper, we propose SDN-based traffic engineering approach that combines in one control-plane link utilization monitoring, delay-aware path selection and proactive failure recovery. The formulated problem is defined as the multi-objective path assignment problem in graph G = (V, E). We propose a greedy congestion and latency aware routing algorithm and a mechanism of pre-computation of backup paths. The framework is instantiated as a discrete-event simulator using an Internet2-like backbone topology with 12 nodes, and it is benchmarked against shortest path static routing (OSPF) and Equal-Cost Multi-Path (ECMP) routing over a range of offered loads, single-link failure cases, and scalability tests in larger topologies ranging to 64 nodes. The framework lowers the maximum link usage by up to 44% when compared with OSPF, avoids loss due to congestion for higher loads of traffic at the bottleneck node, and lowers failure recovery time from around 41 ms to 4 ms while also lowering the average stretch of backup routes from 1.59x to 1.21x. We analyze the resulting tradeoffs between congestion, latency, and resilience, scalability challenges in the centralized framework, and future extensions to the framework.