How to automate business intent with an SD-WAN
Automation in the WAN is a powerful tool for an enterprise. It is one of the revolutionary ideas embodied in software-defined wide area network (SD-WAN).
The move to cloud applications is forcing businesses to rethink their WANs. And the inherent complexity in legacy WANs is not making it easier for network managers.
For instance, an assumption embedded firmly in the legacy WAN is that configuration happens at the device level—and that networking professionals must focus their skills and time on learning and applying complicated CLI commands, device by device.
But CIOs—and their enterprises—do not really care about individual network devices. What they care about is application connectivity: connecting applications to users, consistently, in accordance with the SLAs for quality and reliability that reflect business-level intent.
The job of an SD-WAN
Imagine this: Instead of configuring the network in a device-by-device manner, whether via CLI or GUI, the enterprise simply captures its business-level intent for a new application (or class of applications) with a simple, one-page GUI.
From there a central orchestrator works in conjunction with devices at the network edge to achieve the application connectivity objectives, all without any human configuration of the edge devices. In essence, this is the objective of an SD-WAN.
An SD-WAN requirement: automation
Automation can quickly and consistently configure a network of devices. It eliminates tedious tasks and the inevitable human errors associated with doing 'the same thing' 10, 100, 100 or even 1,000 times.
The truth is, it's rarely precisely 'the same thing.' Although each location may follow a general template, there are differences and exceptions at each location, from things as simple as site-specific IP addressing, to more complex security configuration. These differences have historically made WAN automation quite difficult.
However, if we focus on the idea of business intent—and rethink the devices themselves with an 'orchestration-first' mindset—today's SD-WAN can indeed translate business intent into action.
A basic automation example: ZTP
An SD-WAN solution must support Zero Touch Provisioning (ZTP). To bring a new branch online, on-site personnel simply plug in power and the LAN and WAN connectivity.
With ZTP, the SD-WAN edge device automatically contacts the enterprise's orchestrator, which configures the device automatically in accordance with the business intent policies for the network.
Later if there is a change in business policy, it can be made centrally with the orchestrator, and the edge devices will be reconfigured automatically. At no point does anyone need to log in and configure an individual edge device.
But most enterprises need an SD-WAN that goes beyond these table-stake capabilities.
Implementing business intent: more than just automation
Automation is not good at dealing with anomalies and unpredictable events. Trying to write automation scripts that anticipate all manner of failure scenarios could very quickly turn into a quagmire. A different kind of intelligence is required to complement automation.
An SD-WAN solution should include learning algorithms that continually monitor the quality of all available paths—and intelligently combine error correcting algorithms with packet-by-packet multi-path load balancing—to deliver a consistent user experience, even when the underlying physical networks experience loss and jitter.
The beauty of the business intent concept is that none of the details of these algorithms need matter to the administrator. All the orchestrator needs to capture is the business intent.
The combination of automation with dynamic learning and adaptation capabilities in the edge devices can deliver consistent performance—without being stymied by the limits of automation.
I believe that SD-WAN needs to be more than just software defined—it must become self driving. Machine learning is an essential ingredient for building an SD-WAN that can dynamically adapt to changing network conditions.
