Podcast: Cloud will consume the world, but it's not the cloud you think
[Editor's note: This podcast originally aired on Nov. 21, 2017. Find more episodes on the STACK That podcast page.]
In this episode of STACK That, Byron Reese of Gigaom and Florian Leibert, co-founder and CEO of Mesosphere, talk with Don Duet, president and chief operating officer at Vapor IO, whose team is putting edge clouds into wireless network base stations.
With the rise of the Internet of Things (IoT), mobile, and more, Vapor IO believes the cloud needs to adapt. It is looking to provide individuals with more computing capacity in a localized temporal mode. This means it brings processing closer to where information and data is being generated—in this case, away from highly centralized cloud data centers to much more local, much more distributed, and much more federated cell towers.
Learn more about how Vapor IO does this, along with the benefits and challenges, and what Duet sees as the future of edge computing.
Byron Reese: Hello, everyone. Welcome back to STACK That, brought to you by Hewlett Packard Enterprise. I'm your host, Byron Reese of Gigaom, and I'm here today with Florian Leibert. Flo is the co-founder and CEO of Mesosphere. Mesosphere makes DC/OS, which is the most flexible platform in the world for containerized data-intensive applications.
Byron Reese, Gigaom
Today, we are going to be talking about the cloud. The origin, you know, of the term "the cloud" is lost in the cloud of the past. They have always been used symbolically on diagrams to represent network space. But the term, as we know it, has only been used popularly for about 10 years. But what a 10 years those have been.
Oftentimes, you know, new technology comes on the scene and it's overhyped and underprepared, and its earliest fans are destined to a series of incremental disappointments as the technology, you know, only sort of works. But the cloud, the cloud, that's different. If anything, it arrived on the scene humbly. Just a convenient way to do backups and other such things, but almost immediately it proved itself to be something else entirely. In those 10 short years, the cloud went, kind of metaphorically, from playing gigs at rotary clubs to filling packed stadiums with screaming fans. Few technologies have exceeded expectations so consistently and so rapidly. The cloud now is powerful. The cloud is secure. The cloud is fast. The cloud is inexpensive.
We are now a decade in on the third generation of cloud technology, and now we talk about different cloud techniques. One of these is edge computing. Edge computing is a response. It's an optimization method, which is a response to a world that needs incredibly low latency. So to talk to us today about the edge and the cloud, we have Don Duet. After a long and illustrious career as the CIO of Goldman Sachs, Don decided to take on an entirely new challenge, a startup, as the president and COO of Vapor IO, and he's starting from the ground up, bringing a new kind of cloud computing service to the market. Welcome to the show, Don.
Don Duet: Well, thank you very much. Very happy to be here.
Don Duet, Vapor IO
Reese: OK, before we get into the technical stuff, Don, I have got to ask: How does the guy who ran innovation, ran IT at Goldman Sachs for all those years, decide to do a startup at that point in your career? Was this something like a bar bet you lost or something?
Duet: Well, it wasn't exactly that. But I would say that as I was working at Goldman…in some ways, and we'll get into this in the dialogue, we think it's very much a…probably the best implementation of cloud and edge computing today, just given the nature of financial markets and the way that it's evolved into a fully digital paradigm. And so my expertise and background had been quite associated, but more importantly, it really was about getting to where I believe the puck is going and where I think are most of the major investments and a lot of the business innovation, when you think about how business meets technology. We think that that's going to happen structurally, much more at the edge. And that's going to drive the need for really important components of infrastructure to be built. And to bring innovation to that table was just a compelling opportunity for me at this stage of my life. I've been at Goldman for 28 years by the way, so I've been there for quite a long run.
Reese: And had you had kind of a culture shock as like the employees show up, you know, with mohawks, and their dog, and barefoot? I'm assuming it's kind of a different feel than it was at Goldman.
Duet: Well, at Goldman, we always said we were a tech company masquerading as a bank. And so we had a lot of, you know, really incredible, great human capital. And frankly, within Goldman, a lot of my personal time was spent really helping lead innovation and drive our own initiatives and next-generation architecture. So it wasn't as big of a step. But it's been fun. It's been interesting. It's been exciting, for sure. And it's been really, you know, having everything to do and not that many people to do it with has created a lot of interesting moments for me. But it's been very, very interesting and exciting so far.
Florian Leibert: That's awesome. So, Don, tell us a little about Vapor IO. As far as I understand, you're actually moving the cloud to cell towers? I mean, aren't you, when you do this, really deconstructing the whole notion of a centralized cloud?
Florian Leibert, Mesosphere
Duet: Well, exactly. That's pretty much…you've kind of nailed it. I mean when we think about the cloud, and we think about the way that the cloud itself is going to need to adapt over the next several years to really meet the localization needs of business, to be able to bend, if you would, to the fundamental aspects of the speed of light. So in order to provide more computing capacity in a localized temporal mode, so you can really do things fast and do them quick and get them closer to where information and data is being generated, we believe that the cloud needs to adapt. That means it needs to move from being a highly centralized collection of data centers that's up in the clouds to becoming much more local, much more distributed, and much more federated.
And so the cell towers are a great example of what we think is the demarcation zone, where the wireless world ends and the wired world begins. And we think that's where fundamentally computing needs to come to, and in particular, what we think of as data center standard computing. So not specialized computing, not computing that puts enormous constraints on the things you can do, but the same kind of computing you think about when you go to AWS or Google or Azure, or any of your colocation providers. So that's very much how we think about the world, building out a meshed-distributed architecture that allows us to bring computing to, as close to, the edge of the wireless and the wired networks as possible.
Leibert: That's really interesting. Can you give us a little bit more detail around how it looks at the edge? What is a cell tower server or maybe even a rack look like? And where do they fit?
Duet: Yeah, that's a great question. When you get to the edge, one thing you'll notice very quickly is that it's not the data center you even thought about. I mean the buildings themselves are fundamentally, substantially smaller. So space becomes a key constraint. So where that is the prime real estate, if you would, it's the beachfront property of the next generation of digital. It's also very constrained in terms of its size, its form factor. It's also very disparate. You know you're working with Crown Castle, who is one of our partners. They have over 40,000 cell towers in the U.S. alone. And again it's a base of…the cell towers for the most part, they're effectively buildings or shelters, which can be leveraged and turned into computing facilities. But space becomes the first issue that you notice that's very different.
The second issue that we are continuing to see is that the edge is not just going to be about the facility but about the interconnect. And so it's really going to be about providing for the ability to have a meshed network to be delivered via fiber. It could be costly in providing the peering. And the things we think about is like meeting rooms in our data centers, but bringing them right up front, bringing them right to where the signal terminates from the wireless network into all the different aspects of the data center and the background network center. So we see that also being a very key aspect of being at this junction point between wired and wireless networks, posting and computing, but also really enabling the network capacity and the network peering to occur, where people can get to disparate content and disparate services. And again, at a much, much more latent perspective.
So, if you think about today, if you're coming in through a device on a mobile network, coming down the cell tower, going now through all the different aspects across the call to get to your cloud data center, that could take somewhere between 250 to 300 milliseconds before you get your information back to your application. As you bring computing and network peering directly to the edge, we can start making that happen in 5 milliseconds. So a massively different performance, a massively different way of thinking about how you can now begin to develop edge-aware or applications at the edge that can really provide both business as well as other forms of just really rapid computing. Because you're taking out a lot of that transient, back-haul capability. And you're really providing that directly up front at the edge.
Leibert: So, how many…that's super-interesting, but how many cell towers exist in the U.S.? And how many servers could that host?
Duet: Well, the cell tower number, I don't have a number off the top of my head. Again, we're working with a number of different owners of specialized property. So that includes people like Crown who own cell towers, companies like American Tower. And again, you're going to be measuring that in, probably certainly, the hundreds of thousands. But there's also other forms of specialized property owners. We're talking to a number of people in the utility space so owners of the distributed electrical grid, who have again, not just power distribution, but they also have physical property. And all these become quite interesting for the edge.
So we see the edge less of something that's been built for purpose, like going in and saying, "Yeah, I'm going to pick these sets of response and start building a greenfield data center. We really see the edge as being created by taking existing footprints, whether they be existing assets like fiber, existing assets like different properties, and then really enabling them to become effectively part of the critical infrastructure by bringing the information, the architecture, the IT, and the services that we provide, through Vapor IO, to really make that an integrated experience—no different than what you're going to find at a major cloud company.
Reese: What is it? You know I just want to take a step back and ask: What's different about how we use computers today that's different than, say, five years ago? Because now all of a sudden you're talking about, like, how sluggish the speed of light is. And so what are we doing differently today that necessitates this rethinking of cloud architecture?
Duet: Yeah, so it's a great question. And what we see—and again, we think of this all as evolutionary—we see this as the continued progression and use of technology in both everything that we do in our lives and…as just pulling, again, more of the capabilities, more interactions into paradigms that are being technologically delivered. And as that happens, then it becomes increasingly necessary for that to be done fast or being done in a forum that enables business versus disables it.
Yeah, secondly, there's a tremendous amount of innovation that's been occurring in the network space, which a lot of what we're talking about today really rests on as we move toward 5G infrastructure, as we move towards the next generation of more of the underlined network for the wireless infrastructure. It's really underpinned by the whole thing. One is becoming exceptionally more... Secondly, it's becoming so that you can do things like packet handoffs, at very different layers, with an infrastructure versus having to go all the way back to a termination point that could be hundreds of miles away. So there's a whole degree of fundamental change, and the technology that's being deployed, to basically deliver the wireless networks that's enabling this degree of adaptivity that we've never had before.
And lastly, which I think is extremely exciting, it's just this whole shift that's been occurring around the world, of what really we call the machine-to-machine usage of the wireless network, where in the past, people like you and I would be the power users of the wireless network. And again, going forward whether autonomic vehicles, whether it be drones, whether it be all forms of industrial IRT, we're seeing that computers are now sitting on the wireless side talking to computers on the wired side. And again that fundamentally changes the equation of terms of the interface, the amount of bandwidth, the performance characteristics. You know humans can only work and speak and read at a certain clock rate. And computers, as we know, are going to take that substantially and exponentially higher.
So these are all factors that are driving the value of digital and the value of business, becoming more and more, requiring that localization of computing data in a way that just performs faster in order to meet these needs of the future.
Reese: And so a question on security. Because your website says there's 200,000 towers, and it's counting things like billboards, doesn't that just multiply the security concerns by 200,000? Like, what keeps somebody from just climbing up the side of the billboard, unbolting that localized copy of the data, and carting it off, and utilizing it for nefarious purposes?
Duet: Yeah, security is going to be a, you know, continuous and a hugely important part of really getting edge computing right, both in terms of logical security, so I'm showing that information is protected, at rest, and in transit, as well as just the physical security elements. One of the things we certainly see, you know, just from the surface-area prospective, is that you have more and more effective data or cloud or cloud data centers, as you just look at the denominator effect of going from the hundreds that probably exist today in the country to hundreds of thousands over time. That's going to require really even much more of attention depth, a much more integrated set of architecture that facilitates that. And so part of the way we see the security paradigm evolving is both along the lines of getting better and more increasing standards for logical security that will become, not necessarily today…perhaps they're almost best practice, and we believe tomorrow they'll become absolutely required.
And then on the physical side, we really believe that the way that physical security will become over time is well, how do you begin to integrate which software, and through software and workflow, to provide for different forms of security? How do you emulate what a security guard would do at a data center, in an unmanned, you know, unpopulated, cloud-edge data center? And so we think those are all really about decomposing the fundamental functions, identity verification, and then recomposing them through software, and through things like AI and computer vision, and so forth, use of cameras, etc. So these are all important aspects of security and will continue to be. I think one of the bigger challenges is those opportunities for the edge.
Reese: Alright, before we continue, I want to take a moment and do a shout-out to Hewlett-Packard Enterprise, who are the people who bring you STACK That. HPE is the leading provider of the next-generation services and solutions that help enterprises and small businesses navigate a rapidly changing technology landscape, like the one we're discussing today. With the industry's most comprehensive portfolios, spanning the cloud to the data center to the Intelligent Edge, HPE helps customers around the world make their operations more efficient, more productive, and more secure.
Leibert: So, Don, you just mentioned the key word: AI. For me, when I think about the edge cloud, I think of like a light cloud, something that's not as powerful as a centralized cloud. What's your thought here? What do you think of the technologies that really make sense at the edge, and that companies, software companies or cloud software providers, have to offer at the edge?
Duet: Yeah, that'd be a great question that I think, that we think of the edge has many different forms, depending on the context that you're looking at. So your car could be the edge, your phone could be the edge, a street lamp posting a piece of computing could be the edge. You know, we also think about the cloud edge, which is where we really believe that the components, or the principles of cloud computing, really get materialized. And so that includes things like assumptions of data center type standards.
So you should be able to bring to that cloud edge, not just a purpose-built device but you should be able to bring a GPU or a classic X86 server. So that's really where we see that intersection between the wired and wireless networks. It's where we believe that the cloud edge will become really kind of very valuable because it will bring both the performance characteristics—not zero latency like the true edge, if the true edge is the car or the phone—but a close proximity ability to do more classical computing like AI.
And I think AI is a huge growth opportunity and growth point. As you think about all the types of things you can do, whether it'll be in societal control systems, things like autonomic vehicles, if you had the ability to do them in the 5 millisecond time frame, the types of problems you could start solving for are radically different from the types you would be doing if you're covering this with a 150 millisecond to 300 millisecond time frame. That's a good…that does require, though, the need for the edge, at least the cloud edge to be able to facilitate for more powerful computing, more standard-type data center-type form factors. And that's really where we're focused on at Vapor.
Reese: You know, it strikes me that Flo's company, Mesosphere, has autonomous car use cases…well, I guess, connected cars technically. Is that something that is kind of at the top of your thinking about the application desk technology? Is that like the uber use case kind of? Uber in the lower "u" there.
Duet: Yes. Absolutely, we think autonomic vehicles are going to be a big use case over time. But we also have been quite excited to see the diversity of the types of use cases at the edge, everything from customer engagement to distribution of content…as people need to bring more and more data that's being generated through IoT, outside of their data centers, into a data center, through to medical use cases.
It's just been a very broad and diverse set of requirements and needs. That again, the ability to balance the capabilities of having data center class-computing very close to the edge of the global and wireless networks becomes a very big positioning aspect that really kind of allows—without having to go back and rethink or refactor the design of the applications—to be able to just bring them forward and actually get tremendous benefit and value from them, as well as, we believe, creating brand new innovations on top of the fact that now you can build applications with no second-type response time.
We think that'll create a whole generation of next-stage of evolution innovations, which is a big part of what we really hope to provide for. The ability to make the edge real from that perspective, for business, for society, really requires that critical infrastructure layer. And that's hopefully what we're going to be helping the world create.
Leibert: So, Peter Levine actually had a really interesting talk recently. So for those of you who don't know Peter, Peter is a general partner at Andreessen Horowitz, one of our investors. And he had a talk titled, "The End of Cloud Computing." Now when I hear you, Don, talk about the edge, it really comes to mind like, what are the big cloud service providers going to do about this development? Are they going to go all in on the edge?
Duet: Well, if you look at most, I think every major cloud provider has already made pretty strategic decisions whether it be Azure Stack from Microsoft, Green Grass from Amazon, you know Google's greatest announcements about providing for, effectively, their software in a form factor that can be brought outside of the cloud. And part of that is to meet the demand of certain aspects of business that are uncomfortable with sharing. But also a big part of that is to enable, we think, the creation of the edge, so that the edge doesn't need to be tied to the physical ownership of the property or the data center facility.
And so that's a key component, we think. And again, we really look at this as being very complementary. We think about this as how the cloud adapts to being able to meet the needs of low-latency-type solutions. We think that there'll be, and there'll always be a huge value equation, from having the back of the cloud to pure density, cost performance, etc.
That will be there, and we'll think of this as complementing existing cloud and enabling to really grow out to the edge and get into the last mile, and enable all these different types of use cases.
Reese: So there are a lot of enterprises who listen to this. And I am curious if you have kind of piratical advice on how to implement this technology. Like what are some of the pitfalls that people will experience? And what are some of the workarounds? What are some other lessons you've learned the hard way?
Duet: I think the hardest part in many cases, many of your listeners, if they're running their own data centers would probably feel like, "Hey. You know, what's so different about this?" Maybe I run a set of colos, or maybe I have a number of different private data centers with some back-end data center networking.
The issue, and what we really focus on and what we think is going to be the important part, is the scale here is just enormous. We were talking before about the number of towers, the continuous latency budget, over the expanse of even just the continental U.S., even outside the country. You know, you start getting into this substantially larger numbers of facilities that will be necessary just to deal with the geographical sprawl and the spread.
And so we really think that what's really important to people is to start planning for or thinking about a future that has, let's say, a much larger amount of computing, that's part of their own network graph of their own. And then start thinking about how do you scale? How do you make that happen? How do you take the processes and the steps that you do not just to provision your facilities, but to manage all the serviceability of them and the supply to lifecycle, all that you believe you need to move you into much more automation, much more a software-defined model, in order to allow that to happen.
One of the one things I always found interesting and compelling about the edge is there's a bit of a conundrum that most people think about the edge. They think about disposability. You know it's like small things can kind of break and go away. Yet the intention in what people will be using the edge for is for some of the most critical applications that people have ever designed. So, therefore, it even has to be even higher, from a risk perspective, from a dial-tone point of view, than even when you think about data centers today.
So we kind of look at that, and the way to solve for that is going to be, over time, through just to get investments in design, investments in software, and really enabling the world to scale without compromising operational risk or security risk.
Reese: Alright. Well, that is a great place to leave it. It's a fascinating topic. And Don, I want to just thank you for taking time out of your busy day.
Leibert: Thank you, Don.
Duet: My pleasure. Thanks very much for having me.
This article/content was written by the individual writer identified and does not necessarily reflect the view of Hewlett Packard Enterprise Company.