From Self-Driving Cars to Self-Driving WANs

I realize that the idea of a ‘Self-driving’ WAN could seem like science fiction.

But if you think of the visions propelling companies like Google, Tesla and Amazon, you can begin to realize how big an impact artificial intelligence (AI) will have in the next few years, both on our personal lives—and the way IT runs enterprises.


Enterprises are already turning to Software-Defined WAN (SD-WAN) solutions to connect employees consistently and securely to applications—whether the applications are in the data center or the cloud.

Automation plays a key role in these current SD-WAN offerings, eliminating many of the repetitive and mundane manual steps required to configure and connect remote offices and branch locations.

But automation has its limitations. Automation is not sufficient to translate high-level business goals or intent into specific actions across the network. And automation is not good at dealing with the many unanticipated situations across production WAN deployments.  These are areas where machine learning and AI can come into play.


It’s instructive to look at what’s happening with self-driving cars.

Each self-driving car has hundreds of sensors that collect information to build a real time model of the environment surrounding the car. Artificial intelligence is applied to determine how the car should react at any given moment. A combination of classic control loops and newer machine learning algorithms work in tandem to achieve a high level goal: driving the car safely from point A to point B.

Furthermore, most implementations supplement the car-level intelligence with fleet-level learning. Every car provides data into a central repository where learning across all the vehicles in the fleet is aggregated and applied.

Using data from fleet-learning brings important advantages. One is building more complete and accurate maps. Another is better identifying hazards and reducing false positives. And, perhaps most importantly, fleet learning provides a way to track and improve the cars’ software performance.

I believe the self-driving WAN will encompass hierarchical learning in the same way. Learning will occur at the network-device level, at the enterprise level, and, for the enterprises that opt-in, learning will occur in aggregate across a “fleet” of many enterprises.

To take the self-driving car analogy one step further, it’s interesting to look at what Google has done with their Waymo cars. They removed all the manual controls with the exception of an emergency stop button. At the insistence of the California DMV, Google apparently reinstated a steering wheel. But otherwise, the primary interface for a self-driving car is high level goal or intent based: safely transporting occupants from point A to point B via the most efficient route.

As we move towards the self-driving WAN, the same kind of transition can happen. Instead of having to understand an alphabet soup of protocols and manual CLI commands applied device-by-device, the WAN will be driven by high-level business intent. The network administrator will be able to focus more on the services the network is intended to deliver, and their impact on the business, and less on the underlying details of how that happens.

At Silver Peak, we believe that automation is just beginning for SD-WAN. We are working hard on finding ways to more effectively translate business intent into action—with an autonomous, adaptive self-driving WAN. In fact, in my next blog I’ll describe a specific instance of a new Silver Peak technology for application classification that embodies some of the ideas I’ve outlined here.

Meanwhile, I invite you to share this blog with any of your colleagues.