HN834: HPE Self-Driving Networks: The Next Evolution of Network Automation (Sponsored)

Ethan Banks (0:00)

Welcome to Heavy Networking from The Packet Pushers Podcast Network. I'm Ethan Banks, along with Drew, Connery Murray. You can follow us on LinkedIn, and chat with us on The Packet Pushers Community Slack group, and please do. We're chatting with Sunalini, who suggested I not even try to pronounce her last name, and Jose Tellado from HPE Today. And we're discussing the self-driving network. And guys, self-driving is a term we've heard from HPE and Juniper before for a long time now. So we're going to dive into that because you go on, what does self-driving even mean? Do we mean radically automated? Do we mean AI? Have the packets finally become sentient? Yeah, right, the packets can't think for themselves yet, guys, but we are getting to a place where patterns within network infrastructure can be recognized and the reactions to them automated. And for the purpose of this, this is not yet another AI conversation. It's not a single technology that applies here. In fact, this conversation is not even just about technology. A network that can be automated effectively is one that's been built specifically with automation in mind. And so a self-driving network then is one that's not only automated but has a level of predictability that has been designed into its architecture.

And Sunalini, I'm going to send the first question to you here because we need to understand here in 2026 what self-driving even means, practically speaking.

I don't think we're talking about autonomous networks, but are we? What do we mean self-driving?

Sunalini Sankhavaram (1:24)

I think you just said the right words. It's all about autonomous networking, but let me just take a step back, right? Because the predominant question is, what is a self-driving network? Let me draw an analogy to a self-driving car. For those of you who've sat in a way more that now drives like a cab driver versus a very safe car initially, or for those of you who have the privilege of driving your fully self-driving Tesla vehicles, you literally get in the car, you give your destination, and then you just let it run. It predicts, it anticipates, it reacts, it does everything all by itself to keep you safe and to get you to your destination.

That's exactly what a self-driving network does. A self-driving network based on all the telemetry it collects for every user, every session, every minute is able to now anticipate and identify what is causing a good or bad experience. If it's a bad experience, the sole reason for a self-driving network is to identify, why is that experience bad? For example, this particular call we all are on. Supposing suddenly my video freezes up, but I have a little bit of a jitter and my audio is warbled, you're all wondering why.

If I had a self-driving network, not only would it know autonomously that I'm having a bad few seconds of a call, it would also know what was causing that issue, and it would then remediate that issue automatically without you or me having to ever worry about it ever again.

Ethan Banks (3:00)

That was one of the qualifications I was curious about, remediation. Auto remediation has been one of those.

I don't think I trust my network to do that yet from an engineering perspective. Tell me what you think is wrong, but don't try to fix it because I'm not sure we're there yet. You're saying we are there globally, we're there in certain use cases. What do you mean, Sunalini?

Sunalini Sankhavaram (3:23)

Great question.

When we started on our journey for self-driving, we actually began with saying, let me first of all measure user experience.

Now, what does that mean? What that means is in a network as invisible as Wi-Fi, it's a shared medium, but everybody uses it as a default network to connect. In fact, I'm sure you would all agree, our kids, no matter what the issue is when they say that Snapchat is not working, they're like Wi-Fi sucks, it may not even be a Wi-Fi problem, right? But that being said, that Wi-Fi sucks is what we are going after. Before a user tells me Wi-Fi sucks, I should already know the user is not happy. The KPI of the network is knowing exactly what is going to make a user unhappy. That's the journey we've been on for the last decade. That's where a fundamental aspect comes in of measuring efficacy.

When we came out with the service level experience metrics a decade ago, when Mist was born back in 2016, Bob Friday, our CTO, when Sujay founded our CEO of Mist back then, it was all about why they leave their cushy jobs at the big C company and start this wireless company. They didn't start a wireless company. Their vision was to start a company that understands user experience and solves user experiences. That's what led to starting from scratch a cloud architecture that could ingest elementary to understand and measure all good or bad experience. Now, when you understand a bad experience, you want to make sure it truly is a bad experience. That's where we then started the whole process of what we call AI-driven support. If I am not helping my support team solve customer tickets, how am I going to help my customers solve network problems? If my support team can't use my product, if they can't do the intelligence I have built in my system to solve problems really quickly.