ML-Based SD-WAN Tunnel Failure Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional failure detection in software-defined wide area networks (SD-WANs) is reactive, leading to traffic disruption until failures are detected and rerouted, as it relies on keep-alive mechanisms that only react to failures after they occur, resulting in potential data loss.
Innovation Solution
Implementing a machine learning-based predictive failure detection system that proactively identifies tunnel failures by collecting and analyzing telemetry data from edge devices, using a Bidirectional Forwarding Detection (BFD)-based telemetry request to gather relevant variables and train models for early prediction, allowing for proactive rerouting before actual failures occur.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If reactive failure detection using keep-alive mechanisms is used, then failure detection capability is provided, but traffic disruption occurs until failures are detected and rerouted
Solution Approach 1:
The system performs preliminary actions by collecting telemetry data from multiple sources (interface statistics, queue depths, CPU utilization, BFD packet losses) and training machine learning models in advance to predict tunnel failures before they occur. This allows the network to proactively reroute traffic before actual failures happen, eliminating the reactive delay inherent in traditional keep-alive mechanisms.
Solution Approach 2:
The system implements continuous feedback loops where telemetry data is constantly collected from network devices, fed into machine learning models for analysis, and used to update failure predictions. This feedback mechanism enables the system to learn from historical data and improve prediction accuracy over time, allowing earlier detection and response to potential failures.
2Loss of time
If machine learning-based predictive failure detection is implemented, then traffic disruption is reduced through proactive rerouting, but system complexity increases due to telemetry collection and model training requirements
Solution Approach 1:
The system leverages existing multi-functional network infrastructure components. Telemetry data is collected from standard network device interfaces using existing SNMP and NetFlow capabilities. Machine learning models are trained on aggregated data from multiple tunnels and can be deployed across the entire SD-WAN fabric, making the complex predictive capability universally applicable without requiring separate systems for each tunnel.
Solution Approach 2:
The patent introduces an intermediary telemetry collection and processing layer that sits between the network devices and the failure prediction system. This intermediary aggregates raw telemetry data from multiple sources, performs initial processing and feature extraction, and feeds cleaned data to the machine learning models. This mediator simplifies the overall system architecture by centralizing complex data processing tasks.
3Ease of manufacture
If traditional keep-alive mechanisms are used, then implementation is simple and straightforward, but failure detection precision and recall are limited
Solution Approach 1:
The system transforms the failure detection approach by changing the parameters used for detection. Instead of relying on simple binary keep-alive acknowledgments, the system collects and analyzes multiple continuous parameters including interface statistics, queue depths, CPU utilization, and BFD packet losses. These parameter changes enable more precise and nuanced failure predictions while maintaining implementation feasibility through standard network monitoring capabilities.
Data Source
AI summary
In one embodiment, a device identifies one or more telemetry data variables for use to predict failure of a tunnel in a software-defined wide area network (SD-WAN). The device sends a Bidirectional Forwarding Detection (BFD)-based telemetry request towards a tail-end router of the tunnel that requests the one or more telemetry data variables. The device receives the requested one or more telemetry data variables. The device uses the received one or more telemetry data variables as input to a machine learning-based model, to predict a failure of the tunnel.