Challenges of Managing Edge Computing Systems – Gil Shneorson, Dell Technologies

This is Techron tv. Hey guys, thanks for the throw. We're here with Gil Sch Norsen, who's senior vice President for Edge Computing Portfolio and Solutions at Dell Technologies.

And of course, we're talking about all things related to edge computing. Gil, welcome to the show. Thank you, Michael.

Good to be here. We've all been talking about edge computing for quite some time now, and I think we talk about it more than we do it still, but more folks are moving down the path and it seems like it's a little more challenging than we perhaps anticipated, because frankly, there's just more systems everywhere. There's more instances of software and machines to run.

So you've been at it longer than most of us. What are you seeing in these days? What are organizations encountering and maybe, you know, what do you wish they were doing that they would do differently?

You know, um, there is a combination of a few things happening out there, uh, between the availability of telemetry and sensors and a lot of high quality video and streaming data. Um, that's enabling data collection, um, to the enabling technology such as multi-cloud that allows us to train models and hetero heterogeneous compute, which allows us to put alsos of form factors into the field as well as software defined technologies. And five G, which is, uh, both an enabler for edge computing and a consumer of edge computing.

And you combine it with the fact that people are trying to either be more productive or add to the customer experience, and they're all trying to make more out of the data that they generate at the edge. And so I think that's what's, um, causing this, this perfect storm of a lot of the compute in the world, um, flowing towards, um, the edge from data center, from cloud, uh, for various reasons. It could be cost, it could be, uh, networking bandwidth, it could be latency, it could be compliance.

Um, and, um, you know, the, the growth rates are phenomenal. And you know, I think I read, uh, somewhere that I d c predicts that, um, um, the, the, the storage in edge location is supposed to be around twenty seven hundred and sixty three exabytes by 2026. That's an increase of 895 exabytes in 2022.

So clearly there is a shift of where compute is done, where data is created. Um, and that comes with, um, with many new challenges that we haven't had before. It seems like to me that one of the nuances that gets lost is we are trying to process and analyze more of that data at the point where it's created and consumed at the edge.

So not everything is moving across a network and back into a cloud. And what is moving is more of an aggregate, but it's kind of a new style of computing. Do people really understand or appreciate what it takes to build and run applications and then connect them back to the cloud or a local data center or wherever it might be?

So I think people have appreciation, but what they've been doing forever is buying in a very siloed manner. So they've been buying IT application at the times. The thing about a retailer that buys going off sales and then, you know, a year later as, um, fraud detection, you know, machine, uh, machine vision base, and then the safety and security system is separate and the, um, new, you know, Kio kiosk delivery or curbside delivery is different.

We've been buying in very siloed manner and solving the problem again and again and again. So in a way, the, the, we have been providing solutions as the industry, but we've been doing it in a very siloed manner. And I think more and more people realize that what's happening at the edge is what used to happen at the cloud, which is kind of a shadow it, you buy a lot of those things on a case by case basis, but then you're trying to connect them to the network and you're trying to secure them, and all of a sudden, um, you have many edges.

And this proliferation leads a lot of people to rethink strategies, which is what we are trying to do. We are trying to look at how to do something in a consolidated and architectural way that you can both secure it and simplified, but consolidate all of those use cases so you don't have to do, you know, the same thing over and over and over again. So I think to to, to short answer to your question, I think people, um, understand it.

They've been solving for it in application at a time, and more and more of them are realizing it's time to rethink that strategy. As part of that, do we need to rethink the relationship between the IT folks and a lot of those OT folks that are out there that have been building and deploying a lot of these solutions over the years. And frankly, they don't always get along with each other.

I dunno if they don't get along. But, um, in, for, for the longest time they didn't have to. And what caused that, because it's specifically in industrial where the OT term is very, um, common.

It used to be that we never connected anything to a network. And so you had your OT network, it was completely segregated from the IT network and it was operated separately with different safety protocols and um, with different lifecycle, um, um, considerations. For example, the OT network could have machines that would work for 25 years and you know, rate of change is very, very low and they really don't want to touch it.

Versus the IT network that's connected to the internet, you know, is always exposed to all sorts of security challenges. And so patching and updates is happens almost nightly. Once we started connecting those two worlds, that's where organizations started needing to discuss and talk and strategize.

Um, these days, most industrial environments are connected to the internet for various reasons. And so, um, the OT side of the house is coming to it to say, connected to my, you know, connect to the network, help me secure. And it is now challenged with all of those environments, which is why I said before, it is it that usually thinks about consolidation and architecture, um, and, and creating a, a, a, uh, a common solution across the board.

Um, we're not there yet, by the way. Some organizations are more mature and have a combined team. Some CIOs have a person on their team looking at the other side of the house.

Um, some of them are still very segregated. So, um, but, but the need is is definitely there, as I said, just because everything is now connected. Um, and I think organizations are responding as a, as a second phase to that, uh, that what already is happening around the technology.

Are we seeing any moves towards standardization? Because when I look out at the edge over the years, there's a little bit of everything floating around out there. There's kinda, um, unique processors and really small instances of memory to entire, uh, hyperconverged systems and they're all slightly different.

So do we need to kind of pull that together somehow? I think that's very valid and I think it's gonna remain like that for the most part, um, from a hardware perspective. Um, because we're, we're dealing with environments that are very constrained from a space perspective, power perspective, cooling perspective, um, the ability to protect, you know, machines like dust proofing and shock proofing and vibration.

So in that world, um, there is a lot of commonality because people want to optimize for a very specific use case. Um, the other side of it, what we call the near edge, you know, you see more and more solutions like H C I that are not in a data center, but they're still in a clean room or in a ruggedized cabinet. And so there you see more standardization.

Um, so I think, um, on the near edge where it's a more of more of it, although the, the lines are blurred, but where there is more it-like solutions, you see more common hardware, more common software, more, um, cloud stacks, more a c i stacks on the very far edge. Um, you see much more diversity and I think that's going to, um, stay like that, especially, um, given that, you know, AI inferencing is otherwise. And so I think you're gonna see even more fun factors.

What we're trying to do is, um, come, you know, solve for the management problem. And so what we think is that regardless of the fact that you have very specific optimized hardware, um, you need to be able to manage it in a very similar manner in an automated, simplified, secure way. Um, and so as Dell, we have all of the pieces we need to, we, we do not have all of the form factors that you mentioned, probably never will.

Um, which is, um, partly why we focus our efforts with, uh, specifically with what I'm doing these days around the solving the management problem. I think it's a very common problem regardless of the hardware form factors. What is the appetite for managing all these endpoints among the IT organizations?

Or are they leaning more on managed services or are they looking for some sort of hybrid model where it's a shared responsibility? I mean, how do I do this stuff at scale? So, so with everything in our industry, the answer is always, it depends.

Um, the edge is highly verticalized, right? And so, you know, if you are in industrial environment, you're probably looking for a very specific solution for quality management or operational efficiency. If you are in retail, you're probably looking at how to make the, the experience better for your customers.

Um, and, and so the, the appetite is, is varied. If you're in an industrial environment, it's very unlikely that you're gonna consume something as a managed service because you actually are not interested in more people walking down your factory floor. Um, there is, there's a lot of concern there.

Um, and to be honest with what I said before, with a very low rate of change, um, it's not so much that you wanna manage things, you mostly want to replace them exactly, like for like when they break. If you talk about a retail store, um, there are many, um, deals out there where it's a combination of managed services completely end-to-end, or parts of the different, um, elements of the edge compute are given to different parts, uh, different, um, vendors to manage. And over there you can definitely see managed services.

Um, and so I think those between retail and industry, those are kind of two bookends of the edge market, you know, very maybe call them archetypes of, of customers. And everybody else, I think falls somewhere in between on that spectrum. Is it harder to build and deploy applications on these environments than, say, in the cloud where I kind of have access to within reason almost infinite resources, but at the edge I feel like, you know, there's a certain amount of memory and there's a certain amount of, uh, bandwidth and you gotta be a little more careful.

Not only do you have to be more careful, um, you you need to be extremely planned, but even the mechanics of how do you get this application that you've ostensibly built perfectly to that edge location is not simple. I, if you are in a, an organization that has adopted, you know, C I C D practices and have a DevOps team and can promote your applications into a cloud environment, that's today in most environments fully automated. But what happens when you develop that retail application or a new, um, um, well, I, I, I like that specifically for retail.

'cause there's a lot of innovation happening. What happens when you develop this new value for customers in the store and you want to promote it to all stores, then you could develop it, you could post it, you could possibly test it in specific test environments, and then you stop and then you give it to other people who have to go and use multiple ways to deploy it into all of those stores. Sometimes it's literally a manual process.

Sometimes it's batch processes, but, um, it's not easy. Um, and in fact, I've talked to some, um, customers that by the time they refresh the whole estate, they have to start over. Um, now what happens when you want to do a nightly update of menus?

Let's talk about fast food. You know, there are fast food chains that send, um, the complete menu and, you know, everything in digitize today. Um, every night it changes pricing changes in places, menus change, graphics change.

What happens when you want to update thousands of stores every night? You need very specific processes for that. And so, um, you know, to develop applications to the edge, you have to be conscious of the endpoint.

You don't have internet resources, but you also have to be conscious of how do you get that, um, content that you develop to those edge devices. Um, that's, that's a challenge of its own, Right? Only thing harder than deploying it, of course, is rolling it back.

And the reason you have to roll it back has something to do with compliance and security usually. So what is the state of compliance and security at the edge these days? Are we getting better at it or do you think we're gonna maybe encounter more pain before things actually do get better hopefully someday?

Well, I think, I think, um, related to what we discussed before, um, the, the edge environments are usually not behind a lock-in key, like a data center. So some of the physical ways to protect equipment are just not simply there because it's out there in the wild or where we all live and we can touch it and we can put s b sticks in it and we can disconnect. And so there's many things we can do to, um, those devices.

And even even worse, um, when we have a, a security breach at the edge, it's usually the, the implications could be not just monitor, it could be human safety as well. Think about a, you know, a breach in a water facility right in the edge, compute and contamination of, of, of city water. Or think about a robotic arms moving in the wrong direction where it's supposed to.

And so there is a lot of things that could happen, which is why security is problem probably lam number one for everybody that's trying to introduce more, um, more innovation, uh, at the edge now, for example. And I think many people are working on it, but we have taken a, a specific point to solve for that problem. And so you connected to what you talked about before, Mike.

You send something into an edge device and you know, now you wanna roll back, you said roll back. Um, so con you know, con um, concepts like immutable operating environments, right? The fact that you can roll back set to factory default, um, are not only important for simplicity, but also for security.

And you have to remember that you have to do all of this without any expertise in the field. So you don't have somebody that can go do it for you. It's all gonna be done either remotely or completely automatically by an unqualified, um, operator.

Um, so to your question, whether it's gonna get better, I, I think it's gonna be better because a lot of us are working on it in the iot space or a whole ecosystem of companies, uh, focus on only iot security. Um, but certainly it's a challenge to be sold across the actual application, the hardware and the management infrastructure that you're putting in place. You mentioned using AI at the edge, but let me ask you this.

Are we gonna use AI to manage all these edge computing devices to save us from ourselves? Uh, I think, I think, um, in the long term, it, it must happen. And I know, I don't know if, if to manage all of them completely autonomously, but certainly, um, there is a, especially at the edge, because you have so many devices, there is an opportunity to aggregate data from them, see behavior patterns, and start making decisions based on them.

So I think everybody that is, is in edge management, ultimately will get to some, um, AI in, we call it AI in management practices. So, so you can make those environment even more because as I said before, because the ultimate goal is to simplify the management of all of those devices. So you can consolidate, so you can simplify, so you can get more value out of the data that you're creating.

So I think what you just said is inevitable. So ultimately, what's your best advice to folks as they kind of embrace edge computing? You've seen enough people go down the path already.

What's that one thing you wish, uh, people wouldn't do? Because it makes you shake your head and maybe they could do something better? Uh, you know, I don't know if, if there's anything people do that you know is absolutely wrong, what I would like people to know is get educated about what, what, what they call the, the art of the possible, right?

Just a while ago, you couldn't imagine having, you know, a, a automated, um, deployment for edge devices. In other words, just an ability to connect into network, power them on and have them be onboarded completely touchless. It is now possible.

Um, just a while back, you didn't think that you could manage all of them remotely. Um, that can now happen. And just a while back, you didn't know that there was an, an ability to deploy applications from a catalog just like you do, to a cloud, to all of your edge devices in a, in a, you know, in a batch operation or with a workflow that you've predetermined.

Um, and so because they know those things are now possible, you know, my, my, uh, wish is that people get educated on, on now on what is now possible because that would make them rethink whether they wanna buy every application in silos, or they can do better than that. Um, and I think that's certainly a mission that we have to educate people on what's possible. Uh, as long as they, you know, they, they, they use what they have today and their tools.

I think everybody's doing their best. It's just that they're doing it many times over for every application that they have, which is kind of shameful to, to, uh, to watch. Um, but as I said, you know, the art of the possible is that discussion that we need to have right now.

What can be done. All right folks, well, you heard it here. Everything's kind of moving to the edge one way or another.

It's just now a question of how are we gonna cope with it at scale. Hey Gil, thanks for being on the show. Thank you for, uh, thanks for having me, Mike.

Alright, back to you guys in the studio.