CTO Advisor 099: ESXi on ARM

Machine-generated from the episode audio and not hand-corrected, so names and technical terms may be imperfect. The audio is authoritative.

Hey, how's it going, Mark? I think we're stuck on episode 96 or 97 of the podcast. It's been a long time since we did fresh content, man. It's been a while. I got a job since we talked last, I think. I think you did get a job. I think. It's all going to employment and getting in the way of podcasting. I know, man. If we could all just quit and record podcasts full time, I'm going to be an Instagram model.

That's that's my new goal. Yeah, I'm still working on a hand model thing. It's not you can do it. Oh, yeah. Or a head model. Yeah, that's a good one. That is a good one. But we for our return episode of the new year, we did not disappoint. We got Andre Werkenstein of VMware to join us. We all three now work for VMware. We'll do our best not to talk any NDA stuff on because we had some some pretty good NDA chatter right before the start of the podcast.

But, Andre, go ahead and introduce yourself. Hi, folks. So my name is Andre Werkenstein. I work out of the Boston, Massachusetts office. I'm one of the tech leads on the ESXi arm team. We're doing ESXi arm. You've probably heard about us at VMworld. Let's talk about some of the announcements from VMworld. You know, I was I was excited because I just happened to be in the VMworld VMware office where you guys are located. And we walked by.

You recognized me. You brought me in. You showed me all this cool stuff. One of which is, yeah, it's a bit of a it's a bit of a, you know, of a trick. But I think we'll get into why I think it's important. And that was ESXi running on not only arm, but Raspberry Pis because during the when we first announced the product, everyone was like, yeah, but when will they run on a Raspberry Pi? So just as a side project, you guys made it happen.

Talk about that and some of the other announcements from the show. Yeah. You know, just going back to Vegas, we had the announcement and myself and my colleagues were manning the booth. And so people would come up and I would give them like, you know, the elevator pitch for doing why arm and the edge matters, like edge matters. Look at me like, uh-huh. ESXi. That's the thing that runs on those like the 2U boxes and the 19-inch rack. And you're telling me it doesn't sound very edge-like.

I'm like, well, don't worry about it. You know, we actually have it right here. And I show that the community networking board, the macchiato bin that we ran ESXi on, they're like, sure. But what about the, what about the Pi? You know, so that was interesting insight is that many people don't know what arm is. Everybody knows what the Raspberry Pi is. Moreover, the Raspberry Pi, I think, kind of underlines the entire, you know, the giant maker community around it.

People use it to build the quintessential IoT products. Right. And so I had people come up to me and tell me a little like story around the cool things they do with the Pi. Right. And so, and I'm standing there, I'm thinking like, yeah, yeah, yeah, Pi. Right. And that's the thing that would be really kind of hard to put ESXi on. The kind of thing that, you know, might, there's no real product opportunities from the Pi simply because it's so limited.

But my God, everybody's so excited about the Pi. I have to show something with the Pi. I have to show that I get, we have to show that we get the world right. We understand that this is important to them. We have to take away and learn from that. So we went back into our lab and we worked at it. And I'm happy to say we were able to show ESXi running on the Raspberry Pi. And I can go into why that actually matters.

So, Mark, we're in different forums. And one of the forums that me and you share is the Tech Field Day community. And I don't know if you remember one of the threads in the Tech Field Day community. Those, you know, those guys are pretty hard to please. Yes. And one of the, you know, consistent thing that they said was, why would VMware even do this? Like, what, you can't run a VM on it. So why? And I think it just goes back to the point that people hear ARM and what they hear is Raspberry Pi.

Right. Not an actual use case that they can do with the Edge or integration into other products and all the other stuff. That's kind of where I, in that thread, that's what I said. So, you know, wouldn't mind pulling on that a little bit and see what Andre thinks. Yeah. So, you know, like I said, you know, people hear ARM, they hear ARM, they really think Pi. And they're like, Pi? Can ESXi do something on the Pi? That's a good question.

So if you really wanted to challenge yourself, and the good news is that, yes, you can run ESXi on the Pi. Now, of course, as you all know, ESXi is an enterprise-grade data center product. And it's optimized to run virtual machines. And, of course, if you want to optimize time, that takes space. Right. So if you want to run something really optimized, you know, your memory footprint will be kind of large. Today, ESXi on any architecture takes quite a bit of space.

Right. Now, we heard the planet. We understand. And, in fact, the demo of ESXi on the Pi is very much, it's a promise that VMware is making that we understand that compute at the Edge involves a number of different footprints, completely different footprints. You may range in capability from the very low end to the very high end in terms of compute. You may have different amounts of RAM. You may have different kinds of IOK abilities and accelerators and things like that.

You can run on battery power. Right. And so by showing that, yeah, we run on the Pi, that promises that VMware understands that ESXi needs to evolve with the space as well. Right. In order to unlock the entire ranges of platforms, you know, we've been working hard over the years to make it scale better. Right. So it's important for going to the really big hardware, but it's also important to go on the really small hardware. And so we showed this.

So that's basically it. I mean, of course, if you think about the general kind of thing, we're still working on product definition for the ESXi on ARM. Right. If you think about the general product lifecycle, you know, by the time ESXi on ARM could be a product, the market for one gigabyte platforms, not very large. OK. Right. And, of course, you know, everything, anything can be done technically. Right. You can spend all this time doing something, but then there's the law of diminishing returns.

So is there necessarily something to be done in a product manner on the Pi itself? No. But the Pi is a great forcing function. Right. You aim for the Pi. Right. But then you run fantastically in two gig boards. Right. So the Pi is, you know, one thing I always told my team is that there's no such thing as a useless brain. Andrej, why are we going to run on this laptop or this tablet? It's like, well, you're right, you know, but we're doing this to highlight assumptions we make in our own system.

Right. Anything we assumed. Right. Because those assumptions are probably wrong. Right. Well, at least we should we should see what they are and then figure out a good engineering solution. Right. So with the Pi, we've learned so many things that affect our product planning and engineering roadmap for making ESX work across a range of functionalities. There's a reason why they're called moonshots. I love that you called it a forcing function. If you're able to get it to run on the Pi, then you can definitely get it to run on the two gig board and everything in between, all the way up to, you know, boards with 64 gigs of RAM and 64 processors, blah, blah, blah, blah.

You can do some really interesting stuff. But from a practical sense, let's talk about. This functionality of ESXI being geared towards running VMS and ARM as a platform and kind of where we're starting to rethink the functionality of not just ESXI, but ARM and the data center help help kind of make it real for us. What are some of the potential use cases? OK, OK. So, again, the way we're looking at this, you know, is VMware, of course, is interested in opening up new markets.

Right. We want to grow as a company. It's a whole edge thing. It's a huge green market opportunity. Greenfield market, sorry. And really, it's a paradigm shift where, you know, you have advanced analytics and workloads basically being pushed to places where either the data is produced, right, or the data is consumed. So these are factory floors, oil rigs, buses, schools, TV vans, intersections, you know, traffic lights, anywhere. Right. And, you know, we have all these devices. The Internet of Things is generating so much volume data, you know, you can't just go and centralize the stuff in one place.

Right. We'll run out of the Internet literally. Right. Never mind that frequently you got to act on the stuff, you know, sooner than it will take it to do a round trip to some data center someplace. So if you want to do some kind of, you know, critical workload that monitors and controls, you really got to do it kind of at the edge. And at the edge, what's important is form factor. What's important is being able to run things cool, passively cooled, running them, you know, low power, you know, really not spending much money, both upfront or maintenance wise.

Also, you know, nobody wants to really deploy general purpose hardware because, you know, what's general purpose? Well, a 19 inch rack box is general purpose, but it's not going to live in your Starbucks. Right. It's overkill. Right. So you're going to design hard. You want to be able to design hardware. They'll just have just the things that your workload really needs, your specific use case and not just some random stuff. Yet in this space, ARM has traditionally shined and you're able to produce custom SOCs.

Now, you know what? I don't think this is just an ARM story. Really, it isn't. Right. Well, clearly, you're going to see both X86 and ARM CPUs in the space. Right. So we've got to be ready. Right. So both have to be part of that vision that we're building. All right. So Mark, have you thought through any potential use cases for, you know, these small devices out at the edge at all? I mean, with IOT, when you start looking in like the fintech space, right, the financial tech space or the insurance tech space, you know, if you're doing workers' comp insurance, for example, you might want something that processes a lot of data locally where you're monitoring for accidents and are controlling that via ARM devices.

Or you want to monitor your crop soil to make sure that you're getting the appropriate, you know, pH balance and water saturation so that you can turn that into a profitable business. Collecting all that data is a ton of data and doing that at the edge is, I mean, I think, like I said, about the only way to do that. And then, you know, some of the as I've been as I've talked to Andrea over the past almost year, one of the things that keeps coming into my mindset are, you know, plain sight devices that we have one use case for.

And you think about, well, what happens if you put something with 64 cores and a ton of RAM, but that's also low powered. So, you know, you go into a either real retail space and you see a kiosk. And this kiosk is not just a kiosk, but a little data center at the edge. Whether you're at a mall or even inside of a restaurant, you know, we have these Coke Freestyle machines. Well, what if these Coke Freestyle machines had these serious ARM low power machines on it?

What types of services and analytics type devices or capability could either retailers or restauranteurs provide when they have what's essentially a data, a low power data center in the middle of a very active space? And I think the excellent point there is we need to look at this not from the technical value that we're going to get, because then we'll just be talking about a Raspberry Pi's are cool. Got to look at the actual business value that we bring by going to the ARM form factor and be able to put that.

I love that example, a Coke machine. You know, every time I went to a vending machine at my old job, it never had the right drinks for me. It was always out because they didn't stock enough because they didn't have that deep analytics to say, you know what? These guys and girls drink a lot of Mountain Dew. Maybe we should have one extra row of Mountain Dew. Be able to give that analytics inside that machine and then send it and collect that.

So, yeah, Keith, Mark, you know, you're really on to something. Right. And the thing is, the world that we're building with us does not exist today. It's very easy to think in terms of, oh, well, I'll be consolidating. Oh, yeah. Consolidating what? Like those existing box size devices like with wall wards? Probably not. Right. So very much. We're building a world where we're not moving workloads or we're not consolidating workloads. These will be the workloads of tomorrow.

And customers already want to run these things at the edge. They want to run them at the wind turbine. They want to run them in the TV van or someplace else. Right. These are not. So this is not this is not the you know, you know, like you're not moving something. So you're not, you know, moving. You're not taking something away. Right. To put it someplace else. But this is more. This is a new opportunity.

Right. And you can either do it or, well, you can be left behind. Right. The other thing is in terms of, again, it's not always a consolidation play devices on the low end of the specter spectrum. They may be like a Raspberry Pi. Right. Why? You might only run one work. Why is it interesting? Well, the problem with IOT or in this space is a bit of a science project. Right. So maybe if you're a large company and you're capable of throwing money and engineering, you can kind of solve it for yourself.

Right. But, you know, there are operations problems, you know, lifecycle management, security, isolation, upgrade. Right. How do you upgrade something? So if you break, you know, a thousand devices across the United States, what are you going to do? Nobody wants to be in the news because they created the world's largest botnet either. Right. So, yeah, we saw VMware solve these issues. I mean, not just VMware, but we did in the data center. This is our bread and butter.

But we think this is the major value we can bring to the edge. We can make it easy. I'm sure other vendors can solve each of these problems individually. But then if you have a customer, you know, and they buy from all these multiple vendors, who is going to be left kind of holding the bag? Right. Who is going to be integrating all this stuff together? Will it work? Right. So what we can really provide, right, is we provide, first of all, our virtualization platform allows people to decouple workloads from the actual hardware.

What this means is that they get to pick the hardware that will fit their wallet and their actual workload. It also means they'll be able to switch. Maybe it's ARM today and x86 tomorrow and ARM again, again, depending on what properties you want. On the other hand, the workloads that you run, those don't have to be built against a concrete small box. They'll be working against the virtual, you know, the virtual machine. Yeah, I gave this one customer a bank, this use case for ESXi on ARM.

Just think if you had a ATM and you wanted to do A-B testing of new functionality, like completely new application. And you could easily using, you know, what VMware has done, management tools to deploy these applications or images to your ATM network remotely. And you can roll back, roll forward and roll back is as simple as flipping a switch. Literally, it's what we do with snapshots, what we do with a rolling out, slow rolling applications, all the stuff that we're doing around.

You know, I think a lot of people say, well, why don't you do it with Kubernetes or some other abstraction? And I think that's the point. If you look at VMware and you look at what we're doing with PKS, Heptio, et cetera. You know, I don't want this to be too much of a VMware rah-rah podcast. I think the ideal is that the industry is moving away from hardware based dependencies to abstracting, creating applications, whether you're creating them on a serverless platform, if you're using containers, if you're using virtual machine images.

The idea is to extract the way the programming logic and application from the underlying hardware and ESXi on ARM is just another step in that evolution. It's an IoT transformation. And the other thing that you keep in mind is that the kinds of devices that we're talking about in the end. The end customers, I don't mean the people deploying the hardware, but the people who will see it day to day, like, well, we'll sweep around or something. They won't know it's a computer.

For them, it would be just another appliance, right? Yes, inside of that ends up being a real computer. And that's running workloads, but those are use case specific workloads. And the hardware is optimized for those workloads, really. It's not really general purpose computing, right? So that's really the definition of edge, right? You don't have to sort of imagine what anybody could possibly do, right, given the hardware platform. No, no, it's pretty concrete. We have this customer, we have these requirements, we're going to do this in this box.

The box needs to be on for five years, right? How can we make it as cheap as possible? And how do we provide a common platform that the customer can then, you know, in five years build a new box? You know, maybe they'll modify the requirements and still run the same workloads and manage them as easy. You guys might say, well, that's kind of a pipe dream. Is there anything today in the world that is a little close to this?

And I'll remind you that the demo, the cloud to edge, you know, demo that we showed at VMware Vegas, the wind turbine demo, that was demonstrated on this Marble Macchiato board. It's actually a community network function visualization board. It's a community CPE board. The customer promise equipment place like, and this is literally, you know, Comcast routers, right? Or network switches, stuff like that. The virtual CPE stuff is as close as you can come to this vision, you know, really without abstracting it towards the edge.

The idea that you have a small, passively cooled appliance that internally can run, you know, four to eight different kinds of, you know, virtual workloads. And then be managed, you know, in a distributed sense, right? From across the flat. Pretty much, we want to bring data center technology to the entire planet, tied together. And I think we could talk about this forever, but I don't think we can. Keith, you got any final thoughts? Yeah. Every time I talk to the ESXi team in general, I get, you know, I understand distributed systems a lot better.

Specifically, when I'm talking to the ARM team, it starts to stretch my understanding of compute as a utility. And I think this is what this conversation has been about as we're starting to move away from the data center and thinking through how do we take disciplines from the data center and apply it to the edge. I think we've solved a lot of the problems that exist at the edge. We just need to know how to repurpose those capabilities. And ESXi on ARM is, I think, a great example of that.

Yeah, I agree. I think this is stretching my depth, and I like that. I'd love to, you know, get in a show notes, some, you know, further reading for people or something like that where we can go. So, Andrei, are you on social media or platforms or anything like that? Actually, I run a Twitter account. It's kind of cloaked. It's not that cloaked. I mean, you can tell it's me, but it doesn't have my face on it. It doesn't have my name on it.

It's called, what's it called, What's In Inside? You know, if you can get a look, you can see where that's going. But I do have a Twitter account. I do aggregate ARM community news, because that's obviously relevant to my space. And the things that the public and that we can say about what we're doing, those go on there as well. That's awesome. Keith, where can people find you online if they don't know for some reason? com. And I'm at Sensi Storage on the Twitters.

I say that just to make Keith's family happy. Thanks, guys, and we'll see you next time.