Predictive Tunnel Rerouting for SD-WAN Flapping

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Traditional reactive failure detection in software-defined wide area networks (SD-WANs) leads to traffic disruption until a failure is detected and rerouted, as it relies on reactive mechanisms like keep-alive signals, which do not account for tunnel flapping behavior, where tunnels oscillate between down and up states, affecting network performance.

Innovation Solution

A device in the network uses a machine learning model to predict tunnel flappiness metrics, proactively rerouting traffic from a tunnel expected to flap onto an alternate tunnel before it actually fails, evaluating the performance of the alternate tunnel and adjusting routing decisions based on flappiness metrics and application QoS.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If reactive failure detection mechanisms (keep-alive signals) are used to detect tunnel failures, then failure detection capability is provided, but traffic disruption occurs before rerouting can take place

Engineering Contradiction:

Improvefailure detection capabilityVSAvoidtraffic disruption time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary actions by proactively rerouting traffic before tunnel failures actually occur. Machine learning models analyze tunnel metrics (packet loss, latency, jitter) to predict upcoming failures, allowing the system to switch traffic to alternate tunnels in advance, thus eliminating the traffic disruption time that characterizes reactive approaches

Inventive Principle:

Principle #10Preliminary action

2Device complexity

If simple reactive rerouting is implemented, then implementation complexity is kept low, but network performance degrades due to tunnel flapping behavior

Engineering Contradiction:

Improvererouting system complexityVSAvoidnetwork performance

Core Design Contradiction:

Device complexityVSReliability

Solution Approach 1:

The system implements continuous feedback loops where tunnel metrics are constantly monitored, machine learning models continuously predict flapping behavior, and routing decisions are dynamically adjusted based on predictions. This closed-loop feedback mechanism enables the system to adapt to changing tunnel conditions and avoid flappy tunnels, significantly improving network performance

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system changes routing parameters dynamically based on machine learning predictions. Instead of static routing configurations, the system adjusts route selection, traffic engineering parameters, and tunnel preferences in real-time based on predicted tunnel stability, allowing it to avoid flapping tunnels while maintaining low complexity through automated parameter adjustment

Inventive Principle:

Principle #35Parameter changes

3Loss of time

If machine learning prediction is added to predict tunnel failures, then proactive rerouting capability is achieved, but system complexity increases

Engineering Contradiction:

Improvetraffic disruption timeVSAvoidprediction system complexity

Core Design Contradiction:

Loss of timeVSDevice complexity

Solution Approach 1:

The system introduces machine learning models as intermediary components between raw tunnel metrics and routing decisions. These models act as mediators that process complex metric data and translate it into actionable predictions about tunnel flapping, enabling proactive rerouting without requiring complex decision logic throughout the entire routing system

Inventive Principle:

Principle #24Intermediary (Mediator)

4Speed

If traffic is rerouted onto alternate tunnels without evaluation, then rerouting speed is improved, but network performance may degrade due to selecting suboptimal paths

Engineering Contradiction:

Improvererouting speedVSAvoidalternate tunnel performance

Core Design Contradiction:

SpeedVSReliability

Solution Approach 1:

The system performs preliminary evaluation of alternate tunnels before rerouting traffic. Machine learning models assess the stability and performance characteristics of potential alternate paths in advance, ensuring that traffic is switched to tunnels that are both available and suitable for carrying traffic, thus maintaining performance while achieving fast rerouting

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11343171B2Using a flappiness metric to limit traffic disruption in wide area networks

Publication Date: 2022.05.24 CISCO TECHNOLOGY INC

AI summary

In one embodiment, a device in a network obtains tunnel flappiness metrics associated with a particular tunnel in the network exhibiting flapping. The device makes, based on the tunnel flappiness metrics, a prediction that the particular tunnel is going to flap. The prediction is made using a machine learning model. The device proactively reroutes, based on the prediction, traffic from the particular tunnel onto an alternate tunnel, prior to the particular tunnel flapping. The device evaluates performance of the alternate tunnel, after proactively rerouting the traffic from the particular tunnel onto the alternate tunnel.