Latest / Technology Now / The evolution of self-driving networks: from driver assist to full autonomy | Sunalini Sankhavaram
Transcript
- SUNALINI SANKHARAVAM Wouldn't it be nice if the network already knew you were gonna have a bad experience, knew why you were gonna have a bad experience, fixed that issue automatically so you always have a great experience? That's a self-driving network. SAM JARRELL Okay, so we're back on Self-Driving Networks today, but Michael, we've already done a couple of laps with Self-Driving 1-0-1 this year. What's new now? MICHAEL BIRD Yes, Sam, you're absolutely right. We've talked a few times about self-driving networks and even spoken to people who use them, but I thought it could be nice to revisit the basics behind them very briefly and then go a bit more into the weeds of what's going on within the AI that's running the network. Think of this as like a self-driving, 2-0-1, or maybe it would be 1-0-2. I don't know, Sam. We don't use the same numbering for classes here in the UK SAM JARRELL So we go like 1-0-2, 1-0-3, 1-0-4 until you get to a certain point, and then you go to 2-0-1 MICHAEL BIRD What, agentic gets like 99? SAM JARRELL You don't have to get to all the way to 99. it's a little vibes based. MICHAEL BIRD we should name more things vibes-based, for sure. Anyway, Sam, let's get started. I'm Michael Bird, SAM JARRELL I'm Sam Jarrell MICHAEL BIRD And welcome to Technology Now from HPE. MICHAEL BIRD Okay, Sam, I assume you've come across Marvis before SAM JARRELL Yes, the agent who lives in your network and keeps everything running nice and smooth MICHAEL BIRD That's the one, Because today we aren't just talking about self-driving networks as a hypothetical. we're talking about something that actually exists, and what we're gonna be doing in this episode is digging deep into what is actually going on behind the scenes to keep the network up and running SAM JARRELL Right, because it's a pretty complicated bit of tech. we aren't just talking about autonomy when it comes to self-driving networks. There's the self-healing aspect and the necessary trust of the users MICHAEL BIRD Yeah, and that trust is at the heart of everything, right? but what are the users actually entrusting the network to do? Well, to find out what's going on inside the network itself, I met with Sunalini Sankhavaram, VP Project Management Agentic AI Ops for self-driving full stack enterprise network at HPE. And when we started the interview, she flipped the script a little on me, and I became the one who had to answer some questions SUNALINI SANKHAVARAM Let me ask you a question in return first. Have you ever sat in a self-driving car yet? MICHAEL BIRD No but I have seen them driving past me. SUNALINI SANKHAVARAM And what do you see them do? Um, uh- Do you see a driver in the seat at all? MICHAEL BIRD No driver in the seat. SUNALINI SANKHAVARAM and yet there are people in the back seat MICHAEL BIRD There are people in the back seat. SUNALINI SANKHAVARAM And yet they're going, and they're smiling, and they're on their phones. In MICHAEL BIRD I don't know if they were smiling, but yeah SUNALINI SANKHAVARAM I actually saw a person. I was going to pick up my daughters from their activity, and I saw a car reversing almost into me, and I almost honked, but the car stopped in time, and when it reversed out, there was nobody in the front seat, there was a passenger sitting in the back seat who was on his phone, merrily unaware, and very happy that he was in a self-driving cars But the takeaway for that is the technology is here for cars to be self-driving, where we are trusting our lives to an AI engine and thousands of sensors to proactively preempt what's gonna happen next and then take us safely to our destination. The exact same way, we're bringing that framework to a self-driving network. Now, why does a self-driving network matter? It matters because every day users have problem on the network. You are trying to send an email out of your inbox, and somehow it's stuck in your outbox draft folder. It's not going out, and what do you say? "Oh my God, the network sucks." You're on a Zoom or a Teams call, and there's suddenly jitter, and you lose a few seconds. What do you say? "The network sucks," right? There are myriad such examples where user experience is impacted, and when they escalate to the IT teams, "Oh, it only impacted you for a few seconds, so we'll wait for it to happen again and get packet captures and data and then figure out what's going on," unless the CIO, but then everybody, it's a, all-hand situation. Wouldn't it be nice if the network already knew you were gonna have a bad experience, knew why you were gonna have a bad experience, fixed that issue automatically so you always have a great experience? That's a self-driving network. MICHAEL BIRD Amazing. So, trust is potentially a big hurdle for self-driving networks. this would be the same for self-driving cars. There'll be some people, you're right, who'll be happily will jump in a self-driving car. There might be some people who'd be like, "Hang on, there's no driver. No, no, no, no, no, no, no, no, no, no." So, what are the sort of conversations you're having with customers when it comes to trust about self driving networks SUNALINI SANKHAVARAM Great question, right? Because AI can do a lot of things autonomously, but like you said, you have to trust it to do so. So we have been on this journey of a self-driving network for the last decade, and the focus for us since the very beginning was the efficacy of what we proactively find as problem that are actually impacting users. When we began our journey, when we introduced our first virtual network assistant, Marvis, back in twenty eighteen, we came out with what we call driver assist actions. So just like when you're driving a car, you have blind spot assist and parking assist, the same way we were telling customers, "Hey, you have a bad cable. Hey, you have a missing VLAN. Hey, you have an MTU mismatch or you have insufficient coverage, or you have an ISP issue that's causing a lossy link, causing an end user experience problem." And we developed this whole AI-driven support model where as Bob Friday, our chief AI officer, would tell us, he said support is a proxy for customers. So every ticket coming in on the product, the first thing that the support team does is have Marvis answer that ticket to see how Marvis did. And if Marvis doesn't have the answer, they give Marvis a thumbs down. And that input we take back into the data science team every day, every ticket and say, "Okay, We need to make Marvis smarter and smarter and smarter so that that ticket never comes back in again." 'Cause the minute Marvis recognizes the symptoms, it can fix it automatically, right? That has what has led to decades and decades of us getting the right data in the cloud and training Marvis where customers are now trusting Marvis, right? as we kept building out this high efficacy list of not just alerts, but actual needle in a haystack problems that impact users, there were a lot of those issues where the customer said, "Hey, you know it's a problem. You know it's impacting users. You know how to fix it. Just go fix it." That was the inflection point where we went from recommended actions or driver assist actions to full on self-driving because we had customers telling us, "I trust your AI, I trust the action it's telling me, and I trust you to fix it." MICHAEL BIRD and there are truly self-driving networks out there in the wild SUNALINI SANKHAVARAM There are truly self-driving networks out there in the wild. In fact, the latest one, you may be familiar with the Milan Olympics. That was running on a full-on self-driving networks with agentic AI, where all the issues were preempted. If you speak to the CIO of Milano Cortina, he will say he's never had a better experience, not just from him from an operator perspective, but even from the Olympians who attended the event, who are... As you can imagine, when they're stressed, they're all on social media trying to de-stress. They were all able to do what they needed to do and perform the best possible, MICHAEL BIRD So the idea is, I mean, we've talked about self-driving networks on this podcast before. So the idea is it sort of, self-heals but without any input. Is that right? Yeah. Uh, SUNALINI SANKHAVARAM it self-heals by not having to wait for a human prompt to say, "I have a problem." So the whole point of a self-driving network, and especially now with agentic AI, is all about trying to automatically measure when users are unhappy and then automatically trigger agents to figure out what's making them unhappy, and even better, go fix it. Yeah. But now it's going a step further. If I already know what is a network signature that causes an, a user to be unhappy, how do I fix it even before the user comes on the network? So that's the whole drive towards going from reactive troubleshooting to proactively fixing, to proactively preempting a network problem and fixing it before the user even realizes there was a problem. MICHAEL BIRD I suppose it could be doing things like predicting, user behaviors, even if there might be some obvious user behaviors where it's like, come in on Monday to Friday, 9:00 till 5:00, but actually there might be some weird user behaviors like, three Mondays every year, there's this big spike in traffic, which is really, no one really can track it as a human, but a self-driving network can probably look at that data aggregator to be like, "Yeah, I know I'm gonna have to sort of provision something extra on that." Absolutely. That three Mondays are ... Yeah. SUNALINI SANKHAVARAM are- Yeah. Three Mondays in a month. Whatever it is. But yeah, it's about pattern matching, but it's also about reacting in real time to capacity surges, right? Like you have an all hands that happens maybe once a quarter. So network is designed to deal with the regular traffic, but this once a quarter event puts a lot of drain on that network, and is a network prepared to handle that increased demand of capacity when suddenly you stuff 2,000 people or 5,000 people into one room, which is a typically empty room. And that's where self-driving kicks into the second level, where it's saying, "Okay, a coverage design perspective, I know I'm good, but my network parameters are not optimized. knowing what I'm seeing right now, can I automatically tweak these parameters and make the users happy? Oh, my SLE says right now it's at a pathetic 33%. I can make it 100% if I do this. Let me go do this," and life is good. So not only is it looking at patterns in the past, but it's also reacting to real time telemetry coming in and saying, "This is clearly an anomaly to my baseline and let me adjust dynamically so I can make users happy." MICHAEL BIRD is there also an element of, self-driving security? you know, if there's a big spike in traffic, actually it might be users or it might be something a bit weird that's going on SUNALINI SANKHAVARAM Absolutely, especially if you have our full stack solution and our carrier-grade firewalls. We look at all the user flow data. We know exactly what the policies are for users going to certain destinations, and if we see that there are certain devices misbehaving, good part about that, and this was happening even before agentic AI come along, is that we were able to... We are able to look at the user behavior, analyze their baseline traffic patterns, look at when the anomaly comes up, and automatically quarantine them. So in case it's a malicious user, like a man in the middle, they're off the network. They can't do any more damage. MICHAEL BIRD And, and does that sort of dovetail into the sort of zero trust concept? SUNALINI SANKHAVARAM It does. Yeah. It does, yeah. But it's more than just zero trust because now you're saying zero trust is all about accessing the network securely, knowing who you are, knowing what your device is, that you're authorized to be on the network, and making sure you follow the policies. This is taking zero trust to the next level, where I'm saying I'm continuously enforcing the policy for you. So if I see a violation 'cause a device has gone rogue, let me take it off the network, right? So. MICHAEL BIRD So what does a self-driving network look like to both somebody who works in an IT department and then maybe to a user? SUNALINI SANKHAVARAM So to a user, a self-driving network becomes invisible, because the network is doing its job. To an IT operator, that's a great question, 'cause a lot of people say, "Hey, if the network is self-driving, what am I supposed to do?" They can now do more than do the tedious sustenance and maintenance of a network. They can actually do more projects, be more productive, upskill themselves, do more automation, do phenomenal things with agentic AI that they could never do before, 'cause they never had the time. So let AI do the grunt work. Let's let the humans do the real work of driving more outcomes, versus spending time sustaining the network MICHAEL BIRD So it's letting the people who work in those sorts of environments to do what they're good at and- And upskill and upskill. and actually the stuff that can be automated, the stuff that's, you SUNALINI SANKHAVARAM With high efficacy MICHAEL BIRD the grunt work, 'cause, having briefly worked in an IT department, you're like, you know there are those jobs where it's like you're just literally copying and pasting and pressing enter and, it's just very much SUNALINI SANKHAVARAM the biggest, time consumer for that, is trying to figure out what is the root cause. Yeah. The f- the resolution of the root cause is actually very quick. It could be bouncing a port, it could be shutting a port down, it could be adding a VLAN to a port. So the fixes are easy, but how do you get to that fix? So if AI can help them understand the problem, understand the root cause of the problem, they've already saved time. But now if AI can automatically help them fix the problem, assuming they've given it permission, it saves a lot of time, right? MICHAEL BIRD Do you think we'll ever have a world where all networks will be self-driving or, self-driving networks are just the standard thing? SUNALINI SANKHAVARAM A decade ago, would you have thought self-driving cars would be a reality? MICHAEL BIRD Well I don't know. Uh, maybe, yeah. SUNALINI SANKHAVARAM when I saw all those movies where I would see these cars teleporting and just taking people to where they want, I didn't ever think they would be a reality, right? So if those could become a reality, we're talking about is it trusting our lives, I think self-driving networks can become a reality very soon. That being said, there will always be guardrails. Guardrails are critical. So just like an autonomous vehicle that you give permission to go full-on self-driving, or you, decide what level of self-driving you want, the exact same way, the networks will become self-driving, but the human will always have the opt-in capability to say, "Yep, I want you to self-drive this, but don't touch that. Self-drive this, but don't touch that. Self-drive this to this level, but don't go full-on self-driving." So there'll always be human guardrails around it. But will there be self-driving networks? Absolutely. They're there today. MICHAEL BIRD Will there be self-driving networks? Absolutely. They're there today Uh, will all networks be self-driving, do you think? It SUNALINI SANKHAVARAM It depends on the operators. Yeah. It's all about the efficacy and the trust. MICHAEL BIRD Thank you very much. SUNALINI SANKHAVARAM It was lovely to be here to talk- Thank you, Michael ... about this. Thank you so much for having me. SAM JARRELL You know, that story she told at the very beginning, where, she was trying to pull out, and the other car nearly backed into her, but it stopped just fine. And there was the fella inside the autonomous vehicle just happily, scrolling away on his phone and not paying a bit of attention. It really does underpin that trust is the real story here when it comes to all of these AI-enabled technologies. but especially, from a self-driving component, that once we feel like it works well enough and that we trust it, it is able to really do a lot of things to change our lives. Because I'm sure there could be any number of things that guy was up to. Like, he could have been just mindlessly scrolling or trying to get work done, right? And so I think that we're on the precipice of something really big here MICHAEL BIRD I think what's also really funny was, when Sundalini asked me if I'd seen a self-driving car. I was a bit uncertain because, the only time I've seen a self-driving car was when I was in your neck of the woods in Silicon Valley, and I think they're far more ubiquitous. and so actually, there's probably an element of some people are way more trusting of that technology because they've seen it more and to her point, they've seen it develop from people being sat in the passenger seat driving past them to now self-driving cars. Whereas here in the UK, they're certainly not common. I probably have never seen one on the roads in the UK, So I think there's also an element of it will take a little while for everyone to be at the same level of trusting self-driving networks. There'll be some people that are like, "Yeah, totally get it. I understand it, and I've seen this technology grow." But there'll probably be some people who are still in the mindset of traditional networks, and this self-driving network thing is like, "Oh, what is all that?" SAM JARRELL Yeah, I agree with you. it comes down to like what is your comfortability and how often do you see it? You're right. there are tons of self-driving cars out here. You see them all the time on the roads. the second you go into San Francisco, it is constant. I mean, they're all over the place. And so people here do have a high degree of comfortability with self-driving vehicles. I don't feel like it took very long though, and maybe people are a little bit more eager to also try out these things when they're new. But I, I feel as though, you know, get it in a couple of big major cities and then it'll start to spread further and further and further. And I think similarly with self-driving networks, there's going to be holdout like companies that are like, "Oh, we don't know if we trust having an AI run our network or, trying to manage all of these pieces. It makes us nervous." But you get more and more big name brands behind it, and it's able to help make their day-to-day easier and help evolve the role of IT, then eventually you'll see it become kind of like the thing that you need to have. MICHAEL BIRD yeah, totally. MICHAEL BIRD Now, Sam, we've, we did a self-driving 1-0-1. This was our self-driving 1-0-2. Uh, I don't know, what do you get, like a stamp or a tick or a gold star? Anyway, we've all completed self-driving 1-0-2. Well done, everybody. Uh, but the final question I had for Sunilini was this: if, if we return to this topic at some point in the future, and, you know, we do like a, a self-driving network 1-0-3, we're all back in the classroom again, uh, the blackboard, the chalk's ready for the blackboard, um, what is the next thing that we should be looking for? SUNALINI SANKHAVARAM what's next is do more, So one of the things that we did not talk about is what else does the network do, beyond just providing connectivity. So the network also does a lot in terms of meeting use cases that drive digital outcomes, such as a better shopper experience in a retail store, a better patient experience in hospitals, a better employee experience in a new campus that just opened up. And what are these experiences? When you go to a new office and you have to find the conference room or the nearest printer, you're lost. You're asking people, you're going through the floor plan and trying to figure it out. Wouldn't it be nice if there was an app on your phone that could tell you where to go? So these are things that we do today, but with self-driving they become more and more and more easy where your access point becomes an IoT gateway, and it's self-driving the entire workflow on all the other use cases. SAM JARRELL Okay that brings us to the end of Technology Now for this week. Thank you to our guest, Sunalini Sankhavaram And of course, to our listeners. Thank you so much for joining us. MICHAEL BIRD If you’ve enjoyed this episode, please do let us know – rate and review us wherever you listen to episodes and if you want to get in contact with us, send us an email to technology now AT hpe.com and don’t forget to subscribe so you can listen first every week. Technology Now is hosted by Sam Jarrell and myself, Michael Bird This episode was produced by Harry Lampert and Eva Higginbotham with production support from Alysha Kempson-Taylor, Nik Damarell, Zoe Revis, Beckie Bird, Alissa Mitry, and Jenessa Ayache. Our theme music was composed by Greg Hooper. SAM JARRELL Our social editorial team is Rebecca Wissinger, Judy-Anne Goldman and Jacqueline Green and our social media designers are Alejandra Garcia, and Ambar Maldonado. MICHAEL BIRD Technology Now is a Fresh Air Production for Hewlett Packard Enterprise. (and) we’ll see you next week. Cheers! SAM JARRELL Bye y’all