VMware Telco Strategy with Bryan Liles

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

Hey, you're watching and listening to episode, insert whatever makes you feel good. It's not really important of the CTO advisor hybrid infrastructure podcast. Most importantly, we have Brian Lau, principal engineer in the telco space at VMware. But we'll talk about how the label telco doesn't necessarily matter because he's a principal engineer in VMware. If you're familiar with. Enterprise IT technology companies, a principal engineer is the equivalent of a VP level position, but, quote, unquote, individual contributors. I don't know.

I don't think you don't have any direct reports, right? I actively rebuff them all the time, all the time. So no direct reports, but you have the responsibility of a VP. Before we get into the topic of telco, kind of help help me understand that. How do you have the responsibility of a VP, but not the people or of a VP? OK, well, if we think about what a vice president does, a vice president is generally the facilitator or actually the facilitator, the owner of multiple strategies at a at a large level.

And, you know, that could be in operations. It could be out in consulting land. But in my case, it happens to be engineering and the VP of engineering. What they are responsible for generally is delivery of software across the domain or across a couple of different products. And with that comes management of resources and people. They're not the same thing. I actually operate at the same level. But what I'm thinking about is another few aspects of this, the integration of software at large levels.

And that's large levels inside of VMware, but also large levels outside of the company as well. And then also thinking about the architecture at a large level. And then also my most important job is I'm a VP that is basically put in a position to make good decisions for VMware, not any particular product or any particular person. So sometimes that the principal engineers in this case are are there to make sure that we're doing the best thing for VMware, not particular, not any particular part of the company, something smart that a lot of more companies should do.

Sounds like you have the dream role of any like engineer. Like we always, you know, have this this picture in our head that we'll be at a position that we can make change. But we don't have to actually deal with people. But you have to deal with people. That's a huge part of your job is dealing with people. So there I talked about the good part. The bad part is because I don't have I don't have people. I don't have budget.

So I am I don't I am not. I mean, I have access to budget, but I don't have direct budget. It's not allocated to me. So a lot of my job is I don't rule through Fiat or anything like that. I actually rule through influence. My my goal is to make people think the way that I'm thinking at a large scale so we can move everybody in the same direction. Not by you better do this or you'll lose your job.

You know, same same thing. And another VP would do. But a different tactic. I do think that you need to care and the stick. So I'm the I am the chief carrot officer of my group. I love that. So let's talk a little bit more about your group, which is the Telco group. It is a big deal to have a principal engineer with the responsibility of primarily Telco. So let's talk about what is Telco today at VMware.

All right. So before I do this, let me level set and talk about what is Telco in twenty twenty one. So everyone, if you have a new phone, you can pick a phone up right now and look at it and see that 5G in the corner. And you're like, oh, 5G. It's one more G than 4G. But it's actually more than that. At the transition from the fourth generation Telcos to the fifth generation Telcos. The biggest piece of that was the ability to run these applications that support Telco on commodity hardware.

So before now, you would have to go to a big vendor and they would they would happily sell you a very, very large, expensive stack of custom hardware and software that did all the Telco bits. And with the 5G specifications now, with the introduction of cloud native practices and actually talking and consuming Kubernetes, what we've been able to do now is your your telephone provider. So if you're in the United States, AT&T, Verizon, T-Mobile and soon some others that you will be able to use commodity hardware.

So you could go to Dell or HP or Supermicro and you could actually buy hardware and then you could work with a network provider. So Nokia and Ericsson and Altiostar, Maveneer and other names that you might not have heard of. You can integrate that software yourself and you could actually run. And here's the best part. You could run you could actually support Telco operations in your data center or in the cloud. So this is a whole brand new space. But and this is where VMware comes into the picture.

Is that you know what? If you have to manage a suite of integrations yourself, that becomes very difficult. And VMware over the years has actually created a great base of virtualization and running and then things for running software at scale that allowed us to have Telco, the core part of a Telco network support. So. So let's talk a little history here, because this is really. Interesting. I've heard this line of thought way back, you know, 2013, 2014 in the open stack days that the ability to virtualize Telco resources to enable to offer new, faster services would be improved.

That's still going on. But what was really interesting, your heart, you're kind of reminding me of when we first met, which was at CubeCon San Diego back in 2019. One of the vendors I sat down with talked about using kind of the spare capacity of that infrastructure that you're talking about to deliver low latency 5G type applications. So the Telco could or a service provider could build kind of this virtualized network that could provide the 5G infrastructure services. There would be spare capacity in the service provider, whether it's AT&T or a smaller provider could then sell cloud services on top.

Is that vision still part of the overall modern Telco or is that, you know, kind of. Yeah, it actually is. We're still working on on actually making it be real in the large scale. But let me let me take a step back and explain Telco one more time. And I'll use a little bit more technical terms to help you understand how this can all work. So with the Telco network before, before 5G, you had the radio access network side, the radio, what we call the RAN.

And that part is close to cell towers. So if you see a cell tower thing between a kilometer or two from there, there's a computer attached to it. And that has a backhaul back to some data center somewhere. And that's called the core. And there's where you do basically the core is where in 4G land we would actually have Internet access, all the accounting, everything else in 5G. We actually have that as well. And we have the RAN in the core.

But in the 5G with the with the newer protocols and the newer technologies, one thing that we have is the proliferation of this idea called the MEC, which is mobile edge computing, which means that we can actually take applications that are either latency sensitive or they need a lot more bandwidth or they're low, low energy, which is like the three things, the 5G Telco anyways, and we can put those where they need to be. So if we had we could be close to a cell phone tower for for low latency things for like like cars that drive themselves in the future.

And we could actually take that bandwidth and we could actually slice that piece off and we could push it out to far out to the edge, close to the telcos. And we could have that compute there. It doesn't have to make any round trips back to the core so we can shave off milliseconds of time. That's happening. But the problem is, is that no one is used to thinking about that in the large yet. And that's what we're trying to figure out.

So I think this is where we can all get a little bit confused that as Telco consumers, when I see it by when I see a cell tower, because it's not necessarily a 5G tower. When I see a cell tower, that's probably based on, you know, some legacy technology, 4G and older. I automatically assume that there's a set of services there, but that wouldn't be a correct assumption in the legacy context. There's that that that equipment was and this is my kind of layman knowledge of it, that equipment was rather expensive.

So you would have a tower, but that tower would then have a backhaul to where that 4G infrastructure lived. So just because I saw a tower didn't mean that I have you can kind of all of these modern services I would expect as a consumer. No, that's correct. Actually, even to the day, whenever you see it, it's 5G. So in current and using the current words, you have user equipment and that's your cell phone or whatever talks to a tower within a kilometer or two of that tower is going to be a computer.

And that computer is going to run a piece of software called the distributed unit. And what that does is it basically takes the signal coming off the tower and it basically converts it up the OSI stack. So it goes from layer one, maybe the layer three. And then there's there's other things. There's something called the central unit to see you that actually moves things along. But what we can do now is you don't actually need a tower. You don't need a you don't need a whole bunch of compute next to a cell tower.

You just need that one computer that actually does that translation. And the reason it needs to be within a kilometer or two is because we have not milliseconds. We have microseconds that we have to deal with here and we have speed of light issues. So there is going to be a computer fairly close, but the data is not going to process. It's just actually it's just actually change to a format that is better for moving across networks. So I'm tempted to go in like eight different ways from technology perspectives.

You know, I'm thinking just density of edge compute. I'm thinking about control plane. I'm thinking about the actual hypervisor and technologies that's going into these devices. Thinking about capacity management, all of these questions. Which one of these topics or set of topics are kind of in your purview at VMware? OK, so I'll take a step back and I'll just say, why is VMware, you know, VMware? Oh, the virtualization company. Actually, we do way more than that now. That was like the first chapter of VMware.

The second chapter. You know, we we definitely have things that we are running and we have things like AirWatch and other and other applications. We're running in organizations. Why does VMware care about Telco? Well, you know what? VMware is actually making this new transition. And the writing's on the wall that edge compute is actually, in my opinion, I don't quote me. Actually, quote me on this one. Don't quote the company on this one. Edge compute is going to be bigger than cloud compute, public cloud compute one day.

Why? Because it's more accessible and the barrier of entry one day will be super cheap. And you know what? Why do I need to run computers on somebody else's hardware whenever I can get the compute that I need? Where I need it. Reason we didn't do that before is because it was hard. So I work in the edge part of VMware. And as part of edge solutions, we're thinking of a couple of things. We're thinking of things like zero trust networks.

We're thinking of running workloads wherever they need to be run. But if you squint and you look at it, well, hold on. Telco is a edge is actually edge computing. It's a different type of edge computing. It's Telco edge. But we're running. We're running lots of compute in an inverted fashion. Most of our compute doesn't run in data centers. It runs in inverted. It runs out in the field. And then we're and then we're bringing it back to a smaller core in the data center.

So that's why VMware is doing this, because we actually have a huge interest in edge computing and all from big to small. And with that, you can't just dip your toe in Telco. So you actually have to create a Telco practice to be able to do this. And that's why we've done the investment, because we really want to talk. We want to tackle edge. But to do Telco right, you actually have to make an investment. So I'm struggling with kind of two major different areas because I'm thinking about Telco from the end user air watch type of perspective, controlling and it's not just controlling this enabling devices.

Pat Gelsinger, CEO of Intel, former CEO of VMware. I'm just coming out of the Intel innovation keynote in which the team was talking about that. Each individual user in the future will have basically a teraflop of computing process at their. Disposal, disposal, that's a massive kind of edge challenge when you think of like corporate data and enabling and protecting the data at the edge, zero trust, et cetera. Then I'm thinking about the Telco side of it. There's the movement within one microsecond of that much data within the Telco network.

So, yes, pushing out the the infrastructure to the 5G edge helps with that. But the orchestration of that seems like a really big problem from not just the Telco side, but the end user side. Can you kind of help me tease out my thoughts of where VMware is helping customers focus on both sides, the the edge consumer in the edge provider? All right, so we'll start at the edge provider and how we're looking at this is from a big from like the hundred thousand foot level as we're thinking about automation, orchestration and observability and fault detection and all sorts of things.

So we're thinking about if instead of having all these tendrils of of complexity that are floating around, how can we actually patch it up and in a way that is flexible but still powerful? So that's that's what we're doing in this space. So not to drop my product names, but I have a product called Telco Cloud Automation. And what that does is it helps it actually assist service providers, as we call telcos. And when it comes to installing all the software they need across all the servers they need.

And you can think in some of these telcos, we're thinking of the number of cores is in a seven digits. How do I move software and at that scale to where it needs to be? And then how do I not only do I upgrade the software I upgrade? If we're using Kubernetes, keep that upgraded. If we're using virtualization, keep that upgraded and keep the firmware upgraded, too. And then make it move as it needs to move. And then not only do that, then we need to look at the other side.

How is it working? You know, is this talking to that? Are we getting enough bandwidth? Are there any faults? So these are the products that I work on and thinking at that level. So really what I'm trying to do is provide an ecosystem for service providers to be able to build on and use whatever software they want, but be able to build at the scale they want to with the flexibility want to to get the power they want to and make sure they have the resilience they want to.

So that's from that side. But also from the from the consumer edge side, you know, this is interesting. And this is actually an area ripe for innovation of thinking. We tend to think centralized. We thought centralized data centers. You know what? I'm going to I smile as I say this, but we still think about we talk about AWS and we talk about Azure and we talk about Google. And IBM and Oracle and whomever else we talk about them.

But that's still centralized. We don't talk about putting compute where it needs to be. If I have a let's say if I have a if I have a chain of restaurants and I have point of sale and security there. Why am I actually shipping security footage over the Internet back? Now I got to pay for that. Why don't I just you know what the ubiquity gear I have in my basement can detect events. Why don't I just send events back?

That's what edge computing is going to do for us. Or, you know what, I have a factory that has, you know, 10,000, 100,000, a million devices in it. I just want that thing to take care of itself. Why do I have to ship all that back to a central core? Every bit should be autonomous. And that's where we're working to help actually define what that actually means, because no one literally no one yet has built software to control things at that level.

And that's actually the more experimentation, more experimental type side of what I do. But we're still thinking through, well, what does that mean? What are the best interfaces? And I'll end with this one is that, you know, a lot of people didn't get Kubernetes. What Kubernetes ended up being was a better substrate than Linux for dealing with compute that needs to work across multiple machines. Think about like that. It makes a lot of sense. Forget all the complexity of it.

We don't have that substrate for edge computing yet. And it's not just, you know, building a control pane that actually touches these things. Like, how do we infer? How do we set up networking? How do we set up storage? How do we do faults? How do we do accounting? How do we do security? And that's that's really that's what it all comes down to. What the world that I live in looks like, you know, you remind me of some of the conversations I've had over the past couple of years with some of the large retailers as they were thinking through their digital transformations and enabling applications.

You know, I would challenge them to think about are they thinking about their stores in the right frame? Will the applications that I run in my stores in Wisconsin be the same applications that I run in my stores in California? Should they be the same applications? Should I be thinking about building a infrastructure that allows me to think of my stores as a as a basically a platform as a development platform to to meet the needs of the local market? And even to an extent, you know, you think about kind of the centralized control versus distributed control and ability cannot create regional teams that can take advantage of a common infrastructure, a common platform across all of the footprint.

These are really, I think, heavy things that people need to think about a lot, because, as you said, as you said, it's ripe for innovation and there's not a single vendor. It's more of a community of folks that think about this and getting involved in those conversations back to the telco side of things. So I kind of understand VMware's portfolio enough to understand why the VMware story would make sense to the telcos from a technology perspective. One of the things that you have to deal with as a BP level engineer, you just can't solve a problem from a technology perspective and not think about the cost benefit from a partner perspective.

As you're talking to your telco partners, kind of, you know, let's I know it's way beyond just virtualization, but let's let's take ESXI for an example. Obviously, ESXI would have a lot of benefits in the telco edge, but telco has stayed away from ESXI because of the licensing model. How is VMware thinking about the technology alongside the business challenge of making it work financially for both the telco provider and VMware as a publicly traded company? All right. So I will I will I will dance around this one.

This gets into a touchy area for a lot of people. But, you know, the idea that there is cost associated with running, you know, a non free hypervisor at, you know, many thousands of locations is not something that is unheard of in my space. And, you know, how do how how would I rationalize that? Well, look at this. It's actually easier to rationalize that if you understand if you in the position of being able to have having to support these types of installations.

So if you are a one of the 200 and some odd providers and service providers in the world, you understand that at the end of the day, the cost of ESXI actually is is material. But there's a whole bunch of things that come along with that ESXI and vSphere when it comes to driver integration and things like that. You know, we're almost we're pretty much bar none. Our HCL allows us to understand what works on a particular piece of hardware in any given case better than you having to manage it with bare metal.

And then also with this is that because it's bare metal, because it's not bare metal in this case, you know, now we get into these positions where we can do some more interesting things where we can have flexibility where there was no flexibility and things that might not be provided. And I am definitely being vague here because I'm not sure what I can and cannot talk about. But I will say this. We did demonstrate something at VMware or VMworld. I don't know if it was this year or last.

I think it was this year where there's this new project, Project Monterey. And what Project Monterey is, is this idea where we can take the virtualized piece of the server and put it on the PCI bus. So basically, you know, just take a smaller server and put it in the PCI bus. And now what we can do is you get the best of both worlds. And why would you want that? Well, you know what? There's lots of flexibility and being able to manage hardware with what we bring to virtualization.

But you know what? If you want to eke out all the performance that you need from this from this actual server, we can allow you to do that as well. So back to your question. It's like, you know, what's the value of ESXi? Well, whenever you're whenever so ESXi is just is just the hypervisor. You know, there's dubious value there. But the whole ecosystem, whenever you attach vCenter to it means that you can actually have control of lots of things from a centralized place.

And with some of the plans that I've seen for vCenter, now we can actually build up these constructs where we have managing virtual compute from in a lot of places, and we can actually have a central place for it. You basically turn off that idea that I need to think about every single site individually. And that means upgrading firmware. So basically upgrading firmware BIOS and then operating system. And then what happens to running on that? We can actually take that and we don't have to worry about provisioning things the hard way.

So I'm dancing around this because the great thing I don't work for VMware, so I can I can I can see all kinds of stuff. And yes, you can. And it's OK. But, you know, if you think about it, if you have ESXi all the way to the edge, talking about this driver piece, I can have a really small piece of equipment. And if it's a rule tower, I can have the same services in a rule tower as I can in a big city tower, because I've normalized it, I've abstracted it with ESXi.

So, you know, think of Project Monterey, some PCI bus, FPGA accelerated thing that runs some NSX type features, but on a really low power profile device, something that takes the power footprint of an Intel NUC versus, you know, scaling that all the way up to a system with multiple FPGAs and and several Xeon processors in a really dense area like a stadium that has 5,000 people. And there's a specific application that the stadium wants to run, whether it's security or taking or geofencing, all these cool things you can do when you think about kind of localized 5G.

You need a control plane for that. You need to manage it. You can't scale. You can't do you know, we go into this conversation in the data center. Is it still worth the money to pay for ESXi in the data center versus a KVM? Hypervisor is a hypervisor. Yeah, hypervisor is a hypervisor until you need to scale it and things start to break along the seams and you end up spending people cost, OPEX versus the OPEX or CAPEX when it comes to software licensing.

Software licensing is cheap compared to people. And you're coming out of the Kubernetes cloud native community. You know how hard it is to source that skill set and how important it is to put those resources in the right place. You're a VP of engineering. You know how important it is not to have your smartest peers working on solved problems. So this is you know, this is a solved problem versus an unsolved problem type of discussion and where you want to make your investment in people, I think.

No, I agree. And it is hard to, you know, I never want to come off personally saying, you know, I am. I mean, I advocate for my employer. But, you know, if there's a better solution, actually, this is one of the best parts of my roles. I can advocate for that better solution. Does ESXi work in all cases? No, it shouldn't. No, it doesn't. It doesn't. KVM people who are still using things like Zen. You know, there's things where it doesn't work.

Bare metal can work better. But I will tell you that and I've seen the numbers, the things that we can do with a hypervisor because we have these engineers who know Intel so well, we can give you hypervisor like performance. And I'm not just saying this like as a ad thing. I've seen the numbers. We can give you hypervisor like performance, but still give you the flexibility to basically have that workload being protected by a VM. And that's super important.

And especially in telco, because we use a whole bunch of esoteric things like we're using SCTP instead of TCP and TCP and UDP. In some cases, we use things like precision time protocol, PTP. We're actually doing things like sending GPRS, which is like a which is a packet format for communications. And we're sending that over the network. There's all sorts of low level things like we are super friendly with all these hardware vendors, like all of them. Like I talked to like engineers at Intel, who can tell you about this specific Nick and we're at that level.

So this is the benefit of using virtual virtual machines in some cases, because we can we can actually not just have that that intelligence for us. We can share it with you. Our expertise is your expertise. We bottle it up. And, you know, in telco, you know, certain places, you know, might not work. I was talking about the distributed unit earlier. And that thing is, you know, imagine 40, 50 cores at the at the upper end and use all the memory in a box because, you know, at a cell phone tower, you have many gigabytes or gigabits of data coming out per second.

You know, we're talking at the newer ones. You can go. You can have you know, we're talking at 40 gigabits and that's coming in. Just just coming in all the time. And you have to do this in real time. There's no you can't drop packets. So you know what? There's benefits for virtualization and and we are trying to use those in the best places. But then there's also places where virtualization doesn't doesn't need to be. If you have a small custom piece of hardware and you know what?

We actually have products. And in my business group, we have a we have a sassy or SD Wan product. Fellow cloud. That's KVM. And that's fine. Or, you know what? You can have another smaller product like if you get into like the old like DD WRT where you don't even have that. You could just have this a little tiny device. You know, you might not even need any virtualization on that. You could just probably put Kubernetes on it and be fine.

So Brian, I really appreciate you spending the time to kind of one. Give us the basis of telco modern telco. Because this is something that the enterprise IT industry has been talking about for at least eight or nine years. And secondly, we went a little long compared to our normal podcast, but I think it was worth it. One, we don't get a VP level engineer resource often. That's kind of hard to get. And so I think it was worth it.

I don't know when that's going to be, but I love to kind of get a update on your vision around, you know, this distributed compute problem that we're dealing with. But for people who want to follow you online, how can they do that? I have a LinkedIn page. It's very bare. It has my current job on it. And that's just because, you know what, I'm tired of people coming and asking me about what I did 15 years ago. Just ask me about what I do now.

That's important. So, you know, Brian Lyles, B-R-Y-A-N-L-I-L-E-S on LinkedIn. And then I use Twitter. I don't know. I'm not sure if I'm super effective with it. I like to share things that I find interesting. And so I'm B-R-Y-A-N-L on there. And, you know, I like chatting on there. That's how actually how Keith and I keep up with each other. He'll post something and I'll just come right behind him and say, well, this is what I thought, at least.

Yeah, I've learned an incredible amount from you on Twitter. I think you use it extremely effectively. You post both social issues and technical issues and industry topics that I just love interacting with you. You will learn more about the CTO advisor. You can find us on the web, the CTO advisor dot com four slash podcast to find all the previous podcast episodes. There's you'll even find a conversation with me and Brian from two years ago at KubeCon, where he kind of explained to you.

I credit him for helping me understand Kubernetes, not from a technical perspective, because that part I understood. But he has this really great spiel about the the business value of Kubernetes is well worth the listen. You can find me on Twitter at CTO advisor. DMs are open. You can challenge me. You know, this is not a sponsored podcast, and I think I was asked some pretty tough questions of Brian. You know, the whole pricing thing, he he punted, but that's OK.

If there's other questions you thought I should have asked him, DM me and I'll be more than happy to pass those along. Talk to you next. CTO advisor podcast.