Transcript
Nutanix. Technology moves fast, but sometimes the most important changes are happening way out at the edge of the network. To help us process what's happening there, please welcome Lynn Komp, the head of data center group market readiness at Intel. Lynn, your background spans hardware, software, cloud, which makes you really the ideal person to give us an explanation of what is happening on edge computing and why it's the hottest topic in the data center. Thank you so much, Dan. I am so excited. I love the name of this podcast. It's fabulous. My favorite story about the name of this podcast is that we brainstormed, we used AI, we came up with like a hundred different things, and it was a human brain that came up with I Owe You An Explanation. That's brilliant. I do love it. We're here to talk about something that I have heard a lot about in the past, but I've been hearing more and more about, you've worked in this for a while because you've thought a lot about it and really helped to bring some of the technologies to market that bear on it. We're going to talk about edge computing. Before we dive into any of that, we like to start off with a segment we call the before times. Before times, we will pick a technology and talk about what life was like. For those of you who don't remember, or if you want to take a trip down memory lane for those who do, what life was like. For this time, we're going to talk about ride sharing. Lynn, do you remember life before ride share? Oh my gosh. Vividly. Talk to the listeners a little bit about what life was like and how that had changed once we got it. First of all, you had to find a hotel or somebody that was going to call a cab for you, or you had to find a way to catch the cab going by to get their phone number so you could actually call them. But it was tough. I mean, you would get off in any one of the major airports. You'd be in this huge line. You'd be waiting, sometimes in the rain. And very rarely were you in a city like London where the cabbies are actually adding unique value because they have the entire city memorized. Nope, nope, none of that existed. It was basically raw. Oh, do you remember free credit card pre-ride share? That was great. Do I have enough cash for this? Is there a credit card machine in the cab? Does it work? It was terrible. And then you would get to... I remember my first Uber ride. And the thing that I remember was getting out because when you arrived in a cab, then you had to have a financial transaction. Does the cabbie have change? Do you have enough cash? If not, is there a credit card machine? Isn't going to work? Might be one of those credit card machines. One of those. I remember my first Uber ride. I just arrived and I got out and walked away. And I was like, I am never getting in a cab again. Well, I was in Paris and it was just post-ride share. And here's the irony. The cabbies were just getting ready to protest ride share. And they basically were burning tires around the train station and getting ready to put all the barricades up. And you couldn't get a cab anywhere. And what do I get? I get an Uber in Paris for cheaper. And I remember thinking, and I'm not sure your protest is going to work, guys. Because even though we were trained, don't talk to strangers on the internet and give them your financial information and don't get in cars with strangers. It ended up working great. If you do all those three things the right way, it's magic. If you do everything together, you're fine. Right. And in the beginning, it was also cheaper. That was the incredible part at the time. Right. At the time. All right. Well, that was then. This is now. And maybe we'll even get into vehicles later. But as we dive into edge computing, let's start from the very beginning, that elevator pitch, that 30 seconds to explain to someone who knows what a computer is but doesn't know what edge computing means. How do you get someone started and say, here's how you start framing this problem? Here's one way that I like to help non-techies think of edge computing. If you think about when they're at their favorite Starbucks or their favorite coffee shop, and they need to get Wi-Fi access, and they've got to download the news, they have to be able to get access to their Google calendar that they share with their family. All of that going through that Wi-Fi gateway is a good example of a type of edge computing. It really started with having something that was more local, that was closer in proximity, so that you could reliably get access to information. And a great example of this, going back to remember when. Remember when, on Christmas Eve, your whole family would like to watch a movie, but they couldn't because Amazon Prime was buffering, or Netflix was buffering, and the most popular movies you couldn't get access to. Well, one of the long-standing types of edge computing is content delivery network, or CDN, where essentially they would queue up in these nice, lovely storage environments that are closer to the end users, access to the most popular movies, so you didn't have to wait as long, and you didn't crash the system. So those are all pretty practical. So if you take the case of content, that's a great case. There's a data center somewhere, and maybe several data centers that are actually hosting this. But the edge is not in that data center, it's not in my living room, it's somewhere in between. It's a point of presence, maybe. Is that when I hear people talk about a pop or a point of presence? That's considered edge computing? That is considered one of the edges. And that is the thing that does make edge a little confusing, is there's multiple edges. Because if you think about it, it's multiple domains that are passing off in a handshake information, or holding information that is of high demand and of high use, closer to where it might get used, so you can have it that much faster. You're not dealing with network contention to get access to it. And so the challenge with edge computing is it could be a point of presence, it could be a central office, it could be the edge between an enterprise interchange and the actual internet itself. It could be, I've seen it defined as PC clients are an edge, because they're a point from a connected point, like Wi-Fi gateways, and your client. So what makes edge wonderful is you have your information closer to you, so you can have it faster, you can have it potentially cheaper, with lower latency. The challenge that you have is there's a lot of different edges. Right. So let's talk about a few of those. And I like the idea of talking about a CDN, I think, because everyone can understand that streaming a movie, and you might be streaming something from Amazon, their data center could be, well, let's just say it's the middle of the Amazon, right? It's going to be physically far away from you, literally, like moving those bits is a long way. And if they have it centralized in one location, they might have a billion people trying to hit that server at the same time. So by putting some content at an edge, maybe I'm in San Jose, California right now, so maybe someplace locally, they've got someplace that's an edge location, and they actually keep a copy of that there at that edge location. So when I go to watch that movie, I'm not waiting for a round trip to the Amazon, it's literally a round trip here in San Jose, and that's not serving a billion people, it might be serving the 2,000, 3,000 that are hitting Amazon. That's edge. Yes. Yeah. And in fact, one of the great popular use cases early on was having access to web pages faster. So it was basically caching web pages so that somebody didn't have to wait for it as long to load. These were all when the network wasn't as free, plentiful, and great bandwidth as it is today. But edge computing still continues because if you think of it more like, take a shipping yard, you've got a lot of different cameras for safety, you have different interaction. Sometimes you have autonomous mobile vehicles, like the little Star Wars drones. And all of that safety information, you don't want it going a hundred miles back and then coming in before someone gets alerted that a human stepped into the wrong line of traffic, right? You want it right there so the alerts can happen very quickly. And so you get into things like robotics, edge becomes mission critical. So this is very interesting, and we've kind of moved to a separate use case, and I love it. And that is where you're gathering. So we talked about, you've got a bunch of data somewhere and you're going to put it out there, you're going to cache it so things are faster. But the other thing is happening now, and that's really important. First of all, yes, bandwidth has gotten amazing, but also I've got, you don't want to brag, but I've got a 4K camera in my pocket right now, right? Capable of streaming 4K video, right? And they are literally everywhere. And so if you need to process large amounts of data quickly, you don't have time to do that. So that's another use for edge. Do you have any favorite use cases? The security camera is a good one. Any other favorite use cases for that sort of processing at the edge? You know, one of the things that I've been, that I was really amused with is like a lot of driving examples, traffic management, traffic cams, car cams, right? And somebody told me once that without being able to do analytics at the edge with all of the incoming information from all of the different autos and what was happening during rush hour, think Beijing rush hour, the cameras and all of that information going back raw into the overall highway management system, it looks like a denial of service attack. And so, you know, right? Because there's millions of people that are driving and trying to go to navigate traffic. And so that filtration that what is the really high-priority data and what data do we not need to keep? What can we keep local? What is really important to feed back up in the stream is also something that really helps. And part of the reason is because downstream traffic, so the CDN, I take a movie and then I distribute it, that was configured to be 10 times as much bandwidth and speed as the uplink because people weren't really thinking about uploading video when the networks were designed. And so it's a really important way that you can sift and sort and keep that uplink clean for the information that really matters. It's interesting because it strikes me one of the strike, no pun intended, Major League Baseball I know has clusters in every, not only Major League Stadium, but Minor League Stadium because when you watch a baseball game, they're gathering data for each pitch, visual data, and they want to tell you where did that pitch go? How fast did it go? Was it a fastball, a curveball? They want to tell that in real time. And I don't remember, maybe it's 80 gigs per pitch. Maybe it's more that they gather. They couldn't possibly move all that data, do the analysis in time because you want it on your screen, right? So all that analysis is that. And so I guess that's actually, that's an edge location. Yeah. Hockey games are another good example, following the puck, really understanding where it was, how did it hit? I mean, all of that is an edge application for sure. Right. Absolutely. So I think that the utility of edge is really powerful. Talk to me a little bit about why it's hard. Talk to me about why it's difficult, both from a hardware and software perspective. Why isn't this, I mean, we know how to run computers. We run them in data centers. Is it any different running them when they're not inside the data center? Depending on where that edge location is, that data center could be on the second floor downtown in a historic location. When cell phones were just getting ubiquitous, there were churches in the Northeast of the U.S. that were being able to sell their bell tower to put cell phone towers in their bell towers as part of monthly rental because you couldn't break the historic views by putting a big cell tower in, so you could actually put one in the church clock. So edge locations are difficult because they aren't in this master huge data center that's being maintained, has all of the predictable AC, it has the electricity, it has all the cooling. And then one of the issues that you can run into is truck rolls. Some of the telecommunication service providers, their data centers are in the middle of nowhere. And so you can get out of service and you can have older gear. How much gear do you really need? How much can you power it? If it's not someplace near a water dam at this point, because all these AI data centers are just requiring so much power. And so you're having to make really interesting design choices before you have any idea what the main use case is going to be. And will it last one year or will it last five years? How often do you have to upgrade it? I love the church bell tower, which of course we've all seen in these buildings that are around town. There's literally a church less than a mile from my house in Oakland, California that has those cell towers on it. What's interesting is that, of course, if I have a data center and something goes wrong with a piece of equipment, there are people there 24 hours a day who can go fix that, right? If something goes wrong with...of course, that cell phone antenna is attached to computing. If something goes wrong with that server, there's no one on staff there, right? So you're saying the physical design or how you design that thing has to be different. Even setup, right? I assume setup is very different because you're not going to necessarily have a skilled technician who's right there. So yeah, what do you do about that, about solving that problem? I think that a lot of what I've seen in the industry in telecommunications particularly has been to be really, really careful how the serviceability and the lifespan of the hardware that is in those locations is really going to operate. Think about how often you get your cable or your fiber box updated. Not very often. And so they're having to think in those increments. I once had a telecommunications provider who has an amazing fiber network tell me that their operations people flipped out when they put, this will tell you how long ago it was, 300 watt GPUs in one of their central offices to experiment with some really interesting use cases. And they're like, I can't power this. I can't service this. This thing lasts like it's got the shelf life of a banana. Are you crazy? And so they made them take it out. So there's the power, there's the cooling, there's a lifespan, there's a longevity, there's what applications are you going to run? And communication service providers ended up getting really over the axle when they deployed too much that wasn't flexible enough. So they're looking for what is something that's flexible, it's easy enough to service for the technicians that will give me at least three to five years. And that's hard to find. It's such an interesting problem because especially today, you mentioned GPUs, right? Hardware needs are changing. And if you're going to buy these, the other thing that we haven't really touched on is that when you're building an edge, depending on what kind of fleet you're talking about, you could be talking about thousands, tens of thousands, hundreds of thousands of locations, right? That's right. And so, yeah, you're going to be managing a huge fleet. Clearly you're going to have to do some software updates, but it's not like, oh, just do an install. You're doing 100,000 installs, 100,000 upgrades. Yeah. And data centers are great at that. You know, that remote management lights out data centers that they can just issue patches and updates from one location across millions of servers. You have to think through that when you're looking at sensors, for example. Agricultural sensors and gateways is another really interesting use case where people are monitoring what's happening. I live in Oregon and the Tillamook Creamery monitors what's going on with every cow. And so that's a really interesting edge application as well. Obviously, not every cow has a server bolted to them, but there's a lot of information going back and forth about what do they eat and even like how often did they chew their cud and things like that, because that tells you something about how good the cheese will be. As long as they're still turning out that cheese, they're doing the right thing. That's for sure. All right. So what changes do we see coming in the next several years? You've talked about where we've gotten so far. What do you think is changing in this market? I think a lot of inference is going to be happening at the edge. You know, like one great example, some of the conversations I've been watching about humanoid robots has been how much battery life you can get with the processing that's actually on that node versus how much do you have to offload somewhere else. And that's two different edges, but that interplay of inferencing for emergencies relative to inferencing for the overall management of a humanoid fleet, I think that's going to be really fascinating because how much battery life can we afford? We'd like six hours, please. We're getting two. And so that's going to be a really fascinating physical AI application. We're already seeing a lot of the shipping and inferencing happening at those gateways. Right. That's really cool. Well, thanks a lot. This has really helped out. Now we'd like to move on to our last segment, hot take, warm take, cold take. So I'll mention an opinion, something that I've heard maybe that's going on right now. And you tell me a hot take, like, yeah, you think that's going to happen? Cold take, no way that happens or somewhere in between. I like that you brought up robotics. I'll say I will see a robot walking in my office in the next five years. Hot, cold, or somewhere in between? Well, you work in tech, so I would say hot take. Hot take, you think so? I think you might be right. I think we will see them. I've seen parking lot robots in San Jose. So, you know, those are just a little rollabouts, not walking, but you've already got robotics there. So I think it's a hot take. All right, hot take. All right, well, listen, Lynn, thank you very much for spending some time with us today. Really appreciate it. Very insightful. I think this is one of those terms that people are hearing. And I think that once they listen to this, they'll have a lot more knowledge about what's being talked about. Awesome. Thank you so much for having me. Lynn, thank you so much for joining us. That was really a great look at where the industry is headed and how we got where we are. We will be back here with a new episode once a month on YouTube and your favorite podcast platform. So please do subscribe and we will see you here the next time you need an explanation.
