AI for the Sake of AI? Or the start something else?

Recently I was able to attend TFDx at Juniper’s #NXTWORK conference. One of the new things that Mist was announcing was their ability to identify switching issues based on the AP’s connection to the switch.

Some of you know that I can be somewhat skeptical, and my first impression was that this was just AI for the sake of AI marketing.

HOW HARD CAN IT BE JUST TO FIND A MISSING VLAN?


After all, other vendors have been doing VLAN probing for years to identify this issue. The premise of their demo was the ability to identify a missing VLAN from a switch using AI through the Mist platform.

Check out Mist’s presentation here

It turns out; it’s not that hard. But that simplicity comes with some downsides. False positives are the enemy here and determining if that that VLAN is genuinely missing, is the challenge. Is it really missing or is it just a low-traffic VLAN? What about networks for NAC quarantine? That simple method of determining if a VLAN is present has a bunch of scenarios where it might not be all that effective.

At my day job, we have managed service customers where we monitor and manage their wireless network. And as a wireless-centric company, we may or may not have access to their switching infrastructure. Lack of switching access limits the ability to do things like config templating and auditing.

So how do we turn this on? Activating the feature in my Mist was as simple as doing nothing. Already, the lazy engineer in me is happy. I first configured the trunk, intentionally leaving a needed VLAN and plugged in the AP.

It did take the Mist AI 28 minutes to identify the missing VLAN. Naysayers will claim they could find it quicker, and they would likely be correct. Sure, I could found the missing VLAN by hand, as well as automating VLAN port configs through Python or Ansible. Instead, I went on with my day. And deploying this at scale, I can quickly see how large distributed deployments can benefit from this.

I’m also impressed that it’s a single ML model for all customers. No need to train a model per-customer. With the classification of VLANs into just a handful of types, each customer can benefit from this immediately. So even if you have a routed-access layer network, the ML should be able to classify and identify that VLANs are missing.

While it’s easy to see the benefit of this, I can’t help but wonder where this goes next. Now that Mist has announced that the Juniper EX series switches will be joining APs in the Mist dashboard, what other types of insights will Marvis be able to find. I know ARP is a frequent culprit in wireless networks, as are the plethora of layer2 type features standard on most switches.

While missing VLANs is a useful feature, I expect this is just the start of what’s coming. What are you excited to see?

One comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s