Are you ready for the flood of context-based networking solutions? If not, it’s time to invest in sandbags. After the launch of Cisco’s Intuitive Network solution set at Cisco Live, the rest of the context solutions are coming out to play. Granted, some of them are like Apstra and have been doing this for a while. Others are going to be jumping on the bandwagon of providing a solution that helps with context. But why are we here and why now?

Creating Context

The truth is that we’ve had context in the network for decades now. It’s not a part number that we can order from a vendor. It’s not a command that we type into the CLI to activate. In fact, it’s nothing that you can see at all right now, unless there’s a mirror handy.

The context in networks has been provided by people for as far back as anyone can remember. You do it every day without consciously realizing it. You interpret error messages and disregard those that aren’t important. People know how to program VLANs correctly to segment traffic in certain ways. Security context, application context, and more are delivered by breathing, thinking humans.

We have a massive number of tools to help us create additional context and understand things that are beginning to get out of our control. But these tools are still reliant on a person providing the necessary context to operate. Take red light fatigue, for example. This is a situation in which humans are providing context to a situation being reported. For some cases, the red light means that a condition has passed a threshold and there is a corresponding trigger. However, when the context of that trigger is deemed unimportant, the context applied is “ignore that one”. So much context can be applied to a board full of red lights that we eventually become blind to them and miss a real problem when it pops up.

Humans are great at stretching the bounds of thought to understand why situations need context. Humans know when a link is congested enough to need to be configured for longer routing protocol hello timers. Or when a service policy needs to have a bigger exception for scavenger traffic due to incorrect packet markings. But, the question for the future is “How can humans scale?”

Scaling Context

With the rapid expansion of SDN and programmability, we are quickly seeing that mistakes and errors in context can cause massive issues in a short amount of time. Whether it be a botched upgrade or a script that nullifies interfaces, the system is now capable of making mistakes at a very rapid pace. This sounds like the perfect reason for humans to step into the loop and ensure that mistakes like this can’t happen quickly.

But can humans really scale as fast as needed to keep networks running efficiently? That’s the crux of the issue with modern infrastructure teams. The compute and storage group that is moving toward a DevOps-style frameworks wants to make quick decisions and let the system execute them. When those decisions involve the archaic networking team, the process slows down. Why does the network admin need to put this stuff in manually? Why are they reading through the change orders? Why can’t they just trust us and make it happen?!?

Humans are the current source of context for the network. But, much like many other areas of technology, that context needs to be transferred into a form that can scale. We need to teach the network how to handle exceptions and “think” about how to solve issues. We need to begin the process of training the system to replace us. That’s not because we hope that we will eventually be replaced by a shell script. It’s because we have more important things to apply our knowledge to that need to be solved.

Much like network admins have outgrown the need to manually input VLANs and memorize which ports MySQL needs to have opened on the firewall, so too must we start moving away from constantly checking in on the software running the system to ensure that things are running smoothly. Like the learning television programs that show us what an assembly line looks like when slowed down, we as humans need to let the system operate at the speed it can reach without slowing down to help us understand what it’s doing.

Tom’s Take

I long for the day when I don’t need to look at debug output or error messages and provide my opinion about what might be wrong. Networks with machine learning are still in their infancy. They take way too much processing power to determine issues. But they are getting better. More and more companies are going to start leveraging distributed intelligence to help make low-level decisions about operations. That means humans can start focusing on design matters more. And that’s the kind of context where we shine more than anything else.