Seeing Through the Overlay: Unveiling Potential Bottlenecks in SD-WANs

Software-Defined Wide Area Networks (SD-WAN) offer enhanced flexibility and resilience over traditional WAN architectures by leveraging the Software-Defined Networking (SDN) paradigm. They enable the simultaneous use of multiple WAN connections, potentially across different technologies or service providers, while supporting bandwidth aggregation and improved fault tolerance. However, these connections are abstracted as overlays, concealing the underlying physical network (underlay) from the control plane. As a result, the controller lacks visibility into whether different overlays share critical resources such as links or routers, which can lead to congestion and degraded performance. In this work, we propose an active monitoring framework that classifies overlays as joint or disjoint by analyzing the correlation of latency time series from probe traffic sent across the available overlays. This classification allows the SD-WAN controller to make underlay-aware decisions by selecting only disjoint overlays for routing, thereby minimizing shared bottlenecks. We implement and evaluate our approach in Mininet using a real-world WAN topology. Results indicate that our method achieves accurate overlay classification, leading to substantial gains in throughput, with noticeable reductions in both jitter and latency.