Active probe construction using machine learning for measuring SD-WAN tunnel metrics

1. A method comprising:

applying, by a device, 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, wherein the application traffic is conveyed via a primary tunnel in the SD-WAN;

selecting, by the device, a tunnel in the SD-WAN to probe, wherein the tunnel to probe is different from the primary tunnel via which the application traffic is conveyed and is selected based on its relationship to the primary tunnel via which the application traffic is conveyed;

generating, by the device and based on the cluster, packets that mimic the application traffic; and

probing, by the device, the selected tunnel by sending the generated packets via the tunnel,

wherein applying clustering to the traffic characteristics of the application traffic in the SD-WAN associated with the particular application, to form the cluster of traffic characteristics comprises:

applying a neural network-based model to a time series of the traffic characteristics of the application traffic, to obtain a latent vector representation of the time series; and

applying clustering to the latent vector, to form the cluster.

2. The method as in claim 1 , wherein probing the selected tunnel by sending the generated packets via the tunnel comprises:

determining whether the selected tunnel satisfies a service level agreement (SLA) of the application.

3. The method as in claim 1 , wherein the selected tunnel is a backup tunnel for the primary tunnel via which the application traffic is conveyed.

4. The method as in claim 3 , further comprising:

using a result of the probing as input to a machine learning-based failure prediction model configured to determine whether the application traffic should be rerouted to the backup tunnel prior to a predicted failure of the primary tunnel used to convey the application traffic.

5. The method as in claim 1 , wherein the traffic characteristics are indicative of one or more of: a number of bytes of the application traffic, a number of packets of the application traffic, an average packet size of the application traffic, or a flow duration of the application traffic.

6. The method as in claim 1 , wherein the device comprises a head-end router for the tunnel.

7. An apparatus, comprising:

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

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

a memory configured to store a process executable by the processor, the process when executed configured to:

apply clustering to traffic characteristics of application traffic in the SD-WAN associated with a particular application, to form a cluster of traffic characteristics, wherein the application traffic is conveyed via a primary tunnel in the SD-WAN;

select a tunnel in the SD-WAN to probe, wherein the tunnel to probe is different from the primary tunnel via which the application traffic is conveyed and is selected based on its relationship to the primary tunnel via which the application traffic is conveyed;

generate, based on the cluster, packets that mimic the application traffic; and

probe the selected tunnel by sending the generated packets via the tunnel,

wherein the apparatus applies clustering to the traffic characteristics of the application traffic in the SD-WAN associated with the particular application, to form the cluster of traffic characteristics by:

applying a neural network-based model to a time series of the traffic characteristics of the application traffic, to obtain a latent vector representation of the time series; and

applying clustering to the latent vector, to form the cluster.

8. The apparatus as in claim 7 , wherein the apparatus probes the selected tunnel by sending the generated packets via the tunnel by:

determining whether the selected tunnel satisfies a service level agreement (SLA) of the application.

9. The apparatus as in claim 7 , wherein the selected tunnel is a backup tunnel for the primary tunnel via which the application traffic is conveyed.

10. The apparatus as in claim 9 , wherein the process when executed is further configured to:

use a result of the probing as input to a machine learning-based failure prediction model configured to determine whether the application traffic should be rerouted to the backup tunnel prior to a predicted failure of the primary tunnel used to convey the application traffic.

11. The apparatus as in claim 7 , wherein the traffic characteristics are indicative of one or more of: a number of bytes of the application traffic, a number of packets of the application traffic, an average packet size of the application traffic, or a flow duration of the application traffic.

12. The apparatus as in claim 7 , wherein the apparatus comprises a head-end router of the tunnel.

13. A tangible, non-transitory, computer-readable medium storing program instructions that cause a device in a software-defined wide area network (SD-WAN) to execute a process comprising:

applying, by the device, clustering to traffic characteristics of application traffic in the SD-WAN associated with a particular application, to form a cluster of traffic characteristics, wherein the application traffic is conveyed via a primary tunnel in the SD-WAN;

selecting, by the device, a tunnel in the SD-WAN to probe, wherein the tunnel to probe is different from the primary tunnel via which the application traffic is conveyed and is selected based on its relationship to the primary tunnel via which the application traffic is conveyed;

generating, by the device and based on the cluster, packets that mimic the application traffic; and

probing, by the device, the selected tunnel by sending the generated packets via the tunnel,

wherein applying clustering to the traffic characteristics of the application traffic in the SD-WAN associated with the particular application, to form the cluster of traffic characteristics comprises:

applying a neural network-based model to a time series of the traffic characteristics of the application traffic, to obtain a latent vector representation of the time series; and

applying clustering to the latent vector, to form the cluster.

14. The computer-readable medium as in claim 13 , wherein probing the selected tunnel by sending the generated packets via the tunnel comprises:

determining whether the selected tunnel satisfies a service level agreement (SLA) of the application.

15. The computer-readable medium as in claim 13 , wherein the selected tunnel is a backup tunnel for the primary tunnel via which the application traffic is conveyed.

16. The computer-readable medium as in claim 15 , wherein the process further comprises:

using a result of the probing as input to a machine learning-based failure prediction model configured to determine whether the application traffic should be rerouted to the backup tunnel prior to a predicted failure of the primary tunnel used to convey the application traffic.

17. The computer-readable medium as in claim 13 , wherein the traffic characteristics are indicative of one or more of: a number of bytes of the application traffic, a number of packets of the application traffic, an average packet size of the application traffic, or a flow duration of the application traffic.