Semantic FaultAware (SFA): Proactive Fault Prediction Architecture in SD-WAN with Semantic Knowledge Graphs and Machine Learning

Software-Defined Wide Area Networks (SD-WAN) have emerged as a transformative technology for modern enterprise networks, providing increased flexibility and costefficiency. However, ensuring network reliability and minimizing downtime remain significant challenges in SD-WAN deployments. Traditional fault management methods typically rely on reactive approaches, where network issues are addressed after they occur. In this paper, we present Semantic FaultAware (SFA) architecture to enable proactive fault prediction for SD-WAN, which is a novel framework that leverages semantic knowledge graphs and machine learning techniques. To facilitate a deeper understanding of network behavior and fault patterns, the SFA architecture integrates a semantic layer above the control plane. Moreover, a semantic-based knowledge graph construction approach is introduced to harness better representations of monitoring metric entities and relationships in a network. Finally, we propose graphical embedding techniques to enhance the process of feature engineering and machine learning, thereby enabling the accurate prediction of potential network faults. Experimental evaluations conducted on real-world SD-WAN network monitoring data validate the effectiveness of the SFA architecture in proactive fault prediction. The results demonstrate its superior accuracy enhancement in predicting faults ahead of time, enabling network operators to proactively manage and maintain the network infrastructure.