AI WAN Link Selection to Prevent SD-WAN Thrashing

The patent applies preliminary action by using machine learning models to predict future WAN link stability before actual performance degradation occurs. The system analyzes historical and current link characteristics to forecast stability metrics, allowing the SD-WAN controller to proactively maintain applications on currently stable links even if immediate conditions suggest switching. This prevents the thrashing effect of frequent switches by acting in advance based on predicted future states rather than reacting to current degradation.

The patent implements feedback by continuously monitoring WAN link characteristics and feeding this data into machine learning models that predict future stability. The system uses historical performance data, current link metrics, and predicted future states to make informed decisions. This feedback loop allows the system to distinguish between transient instability and sustained degradation, switching links only when predictions indicate prolonged instability, thereby maintaining system stability while still improving link reliability.

The patent applies parameter changes by transitioning from static threshold-based link selection to dynamic machine learning models that continuously adapt to varying network conditions. The system collects and analyzes multiple parameters including historical link performance, current network metrics, and predicted stability indicators. These models automatically adjust their predictions based on changing patterns in the data, enabling the system to adapt to diverse and evolving network conditions while maintaining a unified implementation framework that builds upon existing SD-WAN architecture.