Predictive routing using machine learning in SD-WANs

1. A method comprising:

instructing, by a device executing a software-defined wide area network (SD-WAN) predictive routing process for an SD-WAN, one or more edge devices in the SD-WAN to report telemetry data to the device at a selected sampling frequency;

obtaining, by the device, the telemetry data from the one or more edge devices according to the selected sampling frequency;

training, by the device and using the telemetry data as training data, a machine learning-based model to predict tunnel failures in the SD-WAN;

after training the machine learning-based model, receiving, at the device, feedback from the one or more edge devices regarding failure predictions made by the machine learning-based model; and

retraining, by the device, the machine learning-based model, based on the feedback received from the one or more edge devices.

2. The method as in claim 1 , wherein obtaining the telemetry data from the one or more edge devices in the SD-WAN comprises:

associating tunnel failures with telemetry variables, to assign measures of predictive power to the telemetry variables; and

selecting one or more of the telemetry variables and the selected sampling frequency, based in part on their associated measures of predictive power.

3. The method as in claim 2 , wherein the selected sampling frequency is selected based in part on a computational load imposed on the one or more edge devices by reporting the telemetry data.

4. The method as in claim 1 , further comprising:

after training the machine learning-based model, deploying, by the device, the machine learning-based model to a particular one of the one or more edge devices, wherein the feedback is indicative of false positives or false negatives by the machine learning-based model.

5. The method as in claim 1 , further comprising:

predicting, by the device and using the machine learning-based model, a tunnel failure of a particular tunnel in the SD-WAN; and

indicating, by the device, the tunnel failure that is predicted to the one or more edge devices, wherein the feedback is indicative of whether the tunnel failure occurred.

6. The method as in claim 5 , wherein the one or more edge devices reroute traffic from the particular tunnel to another tunnel in the SD-WAN, based on the tunnel failure that is predicted.

7. The method as in claim 1 , wherein the device retrains the machine learning-based model until a threshold precision or recall is achieved.

8. The method as in claim 1 , wherein the telemetry data is for a plurality of tunnels in the SD-WAN, the method further comprising:

determining that a tunnel-specific model should be trained for a particular tunnel of the plurality of tunnels; and

training the machine learning-based model to predict failures of the particular tunnel, using the telemetry data for the particular tunnel.

9. The method as in claim 1 , wherein the machine learning-based model is further trained using telemetry data from edge devices in a plurality of other SD-WANs.

10. An apparatus, comprising:

one or more network interfaces to communicate with one or more software-defined wide area networks (SD-WANs);

a processor coupled to the one or more network interfaces and configured to execute one or more processes; and

a memory configured to store an SD-WAN predictive routing process that is executable by the processor, the SD-WAN predictive routing process when executed configured to:

instruct one or more edge devices in an SD-WAN to report telemetry data to a device at a selected sampling frequency;

obtain the telemetry data from the one or more edge devices according to the selected sampling frequency;

train, using the telemetry data as training data, a machine learning-based model to predict tunnel failures in the SD-WAN;

after training the machine learning-based model, receive feedback from the one or more edge devices regarding failure predictions made by the machine learning-based model; and

retrain the machine learning-based model, based on the feedback received from the one or more edge devices,

wherein the apparatus comprises the device.

11. The apparatus as in claim 10 , wherein the apparatus obtains the telemetry data from the one or more edge devices in the SD-WAN by:

associating tunnel failures with telemetry variables, to assign measures of predictive power to the telemetry variables; and

selecting one or more of the telemetry variables and the selected sampling frequency, based in part on their associated measures of predictive power.

12. The apparatus as in claim 11 , wherein the selected sampling frequency is selected based in part on a computational load imposed on the one or more edge devices by reporting the telemetry data.

13. The apparatus as in claim 10 , wherein the SD-WAN predictive routing process when executed is further configured to:

after training the machine learning-based model, deploy the machine learning-based model to a particular one of the one or more edge devices, wherein the feedback is indicative of false positives or false negatives by the machine learning-based model.

14. The apparatus as in claim 10 , wherein the SD-WAN predictive routing process when executed is further configured to:

predict, using the machine learning-based model, a tunnel failure of a particular tunnel in the SD-WAN; and

indicate the tunnel failure that is predicted to the one or more edge devices, wherein the feedback is indicative of whether the tunnel failure occurred.

15. The apparatus as in claim 14 , wherein the one or more edge devices reroute traffic from the particular tunnel to another tunnel in the SD-WAN, based on the tunnel failure that is predicted.

16. The apparatus as in claim 10 , wherein the apparatus retrains the machine learning-based model until a threshold precision or recall is achieved.

17. The apparatus as in claim 10 , wherein the telemetry data is for a plurality of tunnels in the SD-WAN, and wherein the SD-WAN predictive routing process when executed is further configured to:

determine that a tunnel-specific model should be trained for a particular tunnel of the plurality of tunnels; and

train the machine learning-based model to predict failures of the particular tunnel, using the telemetry data for the particular tunnel.

18. The apparatus as in claim 10 , wherein the machine learning-based model is further trained using telemetry data from edge devices in a plurality of other SD-WANs.

19. A tangible, non-transitory, computer-readable medium that stores program instructions causing a device to execute a software-defined wide area network (SD-WAN) predictive routing process comprising:

instructing, by the device, one or more edge devices in an SD-WAN to report telemetry data to the device at a selected sampling frequency;

obtaining, by the device, the telemetry data from the one or more edge devices according to the selected sampling frequency;

training, by the device and using the telemetry data as training data, a machine learning-based model to predict tunnel failures in the SD-WAN;

after training the machine learning-based model, receiving, at the device, feedback from the one or more edge devices regarding failure predictions made by the machine learning-based model; and

retraining, by the device, the machine learning-based model, based on the feedback received from the one or more edge devices.

20. The tangible, non-transitory, computer-readable medium as in claim 19 , further comprising:

predicting, by the device and using the machine learning-based model, a tunnel failure of a particular tunnel in the SD-WAN; and

indicating, by the device, the tunnel failure that is predicted to the one or more edge devices, wherein the feedback is indicative of whether the tunnel failure occurred.