ML-Based Active Probe for SD-WAN Tunnel Metrics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing probing techniques in SD-WANs, such as BFD probes, fail to accurately predict tunnel failures due to their inability to mimic actual application traffic, leading to poor performance in failure prediction models and inefficient routing strategies.
Innovation Solution
The implementation of a machine learning-based approach that applies clustering to traffic characteristics to generate active probes that closely resemble application traffic, allowing for more accurate prediction of tunnel performance and proactive rerouting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional probing techniques (BFD probes) are used to detect tunnel failures, then the implementation is simple and fast, but the failure prediction accuracy is poor because the probes cannot mimic actual application traffic
Solution Approach 1:
The patent creates probe packets that copy the characteristics of actual application traffic by clustering real traffic data and generating synthetic probes that mimic packet sizes, intervals, and patterns. This allows the probing system to achieve accurate failure prediction while maintaining relatively simple implementation, as it replicates essential traffic features without requiring complex analysis systems.
Solution Approach 2:
The system performs preliminary clustering and analysis of application traffic characteristics before generating probes. By pre-processing real traffic data to extract key patterns and storing them as templates, the system prepares accurate probe packets in advance, improving prediction accuracy while avoiding the need for complex real-time analysis during probing operations.
2Reliability
If machine learning techniques are applied to improve failure prediction, then the routing decisions become more accurate, but the system complexity and computational requirements increase significantly
Solution Approach 1:
The patent applies machine learning selectively - using clustering algorithms to analyze traffic patterns and generate probe templates, but not requiring full-scale ML models for every probing operation. This partial application of ML techniques provides sufficient routing reliability while keeping system complexity manageable by avoiding excessive computational overhead in real-time operations.
Solution Approach 2:
The system uses unsupervised learning clustering algorithms that automatically identify traffic patterns without requiring manual labeling or extensive training data preparation. This self-service approach to ML allows the system to improve routing reliability through automated pattern recognition while minimizing the complexity of data preparation and model maintenance.
3Adaptability or versatility
If probes are sent at fixed intervals to monitor tunnel health, then the implementation is simple, but the probes fail to adapt to changing traffic conditions and provide inaccurate performance metrics
Solution Approach 1:
The patent makes the probing system dynamic by clustering actual application traffic to determine variable probe intervals and characteristics based on real traffic patterns. Instead of fixed intervals, the system adapts probe timing and parameters to match actual usage conditions, improving adaptability while maintaining operational simplicity through automated clustering-based configuration.
Solution Approach 2:
The system changes probe parameters (interval, packet size, frequency) based on clustered traffic characteristics. By automatically adjusting these parameters according to observed application patterns, the system achieves high adaptability to changing traffic conditions while keeping operations simple through automated parameter derivation from traffic clustering.
Data Source
AI summary
In one embodiment, a device applies clustering to traffic characteristics of application traffic in a software-defined wide area network (SD-WAN) associated with a particular application, to form a cluster of traffic characteristics. The device selects a tunnel in the SD-WAN to probe. The device generates, based on the cluster, packets that mimic the application traffic. The device probes the selected tunnel by sending the generated packets via the tunnel.