Predicting network states for answering what-if scenario outcomes

1. A method, comprising:

constructing, by a device, a set of controlled what-if input parameters for evaluating a what-if scenario in a network, wherein the set of controlled what-if input parameters relate to features of the network that are independently controlled;

using, by the device, the set of controlled what-if input parameters and state data indicative of a current state of the network as input to a network state model comprising a conditional generative adversarial network that implements a conditioning signal, wherein the conditioning signal a) steers a prediction of values for the state data conditioned on the what-if input parameters and b) is a concatenation of the set of controlled what-if input parameters and the state data;

predicting, by the device, a key performance indicator (KPI) in the network by using predicted values for the state data from the network state model comprising the conditional generative adversarial network as input to a machine learning-based KPI prediction model; and

initiating, by the device, a routing change in the network based in part on the predicted KPI.

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

predicting a failure of a first tunnel in the network, wherein the controlled what-if input parameters relate to rerouting traffic from the first tunnel to a second tunnel in the network.

3. The method as in claim 2 , wherein initiating the routing change in the network based in part on the predicted KPI comprises:

determining that the predicted KPI will satisfy a service level agreement associated with the traffic; and

proactively rerouting the traffic from the first tunnel onto the second tunnel, in advance of the predicted failure of the first tunnel.

4. The method as in claim 1 , wherein the network state model comprises a regression model trained to predict values for the state data conditioned on the what-if input parameters.

5. The method as in claim 1 , wherein the state data is indicative of one or more of: loss, latency, or resource usage in the network.

6. The method as in claim 1 , wherein the set of controlled what-if input parameters are indicative of a traffic for a set of one or more applications being sent between a source and destination in the network over a period of time.

7. The method as in claim 1 , wherein initiating the routing change in the network based in part on the predicted KPI comprises:

providing the predicted KPI to a user interface.

8. The method as in claim 1 , wherein the network is a software-defined wide area network (SD-WAN).

9. An apparatus, comprising:

one or more network interfaces to communicate with a network;

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:

construct a set of controlled what-if input parameters for evaluating a what-if scenario in a network, wherein the set of controlled what-if input parameters relate to features of the network that are independently controlled;

use the set of controlled what-if input parameters and state data indicative of a current state of the network as input to a network state model comprising a conditional generative adversarial network that implements a conditioning signal, wherein the conditioning signal a) steers a prediction of values for the state data conditioned on the what-if input parameters and b) is a concatenation of the set of controlled what-if input parameters and the state data;

predict a key performance indicator (KPI) in the network by using the predicted values for the state data from the network state model comprising the conditional generative adversarial network as input to a machine learning-based KPI prediction model; and

initiate a routing change in the network based in part on the predicted KPI.

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

predict a failure of a first tunnel in the network, wherein the controlled what-if input parameters relate to rerouting traffic from the first tunnel to a second tunnel in the network.

11. The apparatus as in claim 10 , wherein the apparatus initiates the routing change in the network based in part on the predicted KPI by:

determining that the predicted KPI will satisfy a service level agreement associated with the traffic; and

proactively rerouting the traffic from the first tunnel onto the second tunnel, in advance of the predicted failure of the first tunnel.

12. The apparatus as in claim 9 , wherein the network state model comprises a regression model trained to re-estimate the state data conditioned on the what-if input parameters.

13. The apparatus as in claim 9 , wherein the state data is indicative of one or more of: loss, latency, or resource usage in the network.

14. The apparatus as in claim 9 , wherein the set of controlled what-if input parameters are indicative of a traffic for a set of one or more applications being sent between a source and destination in the network over a period of time.

15. The apparatus as in claim 9 , wherein the apparatus initiates the routing change in the network based in part on the predicted KPI by:

providing the predicted KPI to a user interface.

16. The apparatus as in claim 9 , wherein the network is a software-defined wide area network (SD-WAN).

17. A tangible, non-transitory, computer-readable medium storing program instructions that cause a device to execute a process comprising:

constructing, by the device, a set of controlled what-if input parameters for evaluating a what-if scenario in a network, wherein the set of controlled what-if input parameters relate to features of the network that are independently controlled;

using, by the device, the set of controlled what-if input parameters and state data indicative of a current state of the network as input to a network state model comprising a conditional generative adversarial network that implements a conditioning signal, wherein the conditioning signal a) steers a prediction of values for the state data conditioned on the what-if input parameters and b) is a concatenation of the set of controlled what-if input parameters and the state data;

predicting, by the device, a key performance indicator (KPI) in the network by using the predicted values for the state data from the network state model comprising the conditional generative adversarial network as input to a machine learning-based KPI prediction model; and

initiating, by the device, a routing change in the network based in part on the predicted KPI.

18. The computer-readable medium as in claim 17 , wherein the process further comprises:

predicting a failure of a first tunnel in the network, wherein the controlled what-if input parameters relate to rerouting traffic from the first tunnel to a second tunnel in the network.