Today’s episode features an interview between Matt Trifiro and Nick Barcet, Senior Director of Technology Strategy at Red Hat. In this interview, Nick discusses what it means to set edge strategy for Red Hat, where he’s seeing the most edge-related demand right now, why the future of edge must be open, and the floating IP solution he’s constructed in order to get reliable internet while he sails around the world on his boat.
Today’s episode features an interview between Matt Trifiro and Nick Barcet, Senior Director of Technology Strategy at Red Hat.
Nick joined Red Hat in June 2014 and now leads the team that helps decide the future of technologies, currently focusing on Telco, Edge, and AI/ML.
In this interview, Nick discusses what it means to set edge strategy for Red Hat, where he’s seeing the most edge-related demand right now, why the future of edge must be open, and the floating IP solution he’s constructed in order to get reliable internet while he sails around the world on his boat.
Key Quotes:
“When we’re talking about a new field of interest, what people are saying is still not yet normalized. We still have problems with definitions and there is a lot of confusion. If I take the average person in the IT field and ask them, ‘What is the edge?’ The answer is generally linked to mobile phones and 5G. And while these are important components, they are not the beginning nor the end of it. So I think it's super interesting to be in this early phase and trying to bring more clarity to the fog in which we are moving.”
“[Three areas] where we see a lot of edge-related demand right now are: In telco–they are busy deploying 5G, which requires an edge infrastructure for it to be performant. Then automotive–lots of evolution is happening in the auto industry…And then the general industrial environment–whether it's oil and gas or standard factories–is really, really keen on improving their processes by deploying edge infrastructure. But it doesn't stop there. We see opportunities happening in retail, in healthcare, in the public sector, it’s really, really varied.”
“When you're dealing with open-source software, there is never a case where a single company offers that piece of software, there are generally multiple options, and that's another way of protecting you from having to redo everything from scratch. So that's why I think that the edge cannot be fruitful--cannot bring maximum benefit--without being open, because if you are not protected against those risks, you're going to have a very, very costly edge.”
“The evolution of networking…is going to change a lot of things on how we approach the networking problem…The availability of network is going to become a problem of the past. If we look at 20 years from now, we have hope to have instantaneous communication--as in quantum--we could have instantaneous communication from any point to any point.”
“The other evolution that is really interesting is how we are using processing power in more specialized units. The fact that you have DPUs available in SmartNICs and we can use those to do things in parallel to what the main CPU is doing. I believe that this specialization of resources is going to enable a new generation of software that provides for problem-solving quicker and much cheaper than ever before. We've seen that revolution happen through GPU's, but we are only at the beginning of this revolution, and it's going to go through all kinds of discoveries.”
Sponsors
Over the Edge is brought to you by the generous sponsorship of Catchpoint, NetFoundry, Ori Industries, Packet, Seagate, Vapor IO, and Zenlayer.
The featured sponsor of this episode of Over the Edge is Vapor IO, the leader in edge computing. We want to be your solution partner for the New Internet. Learn more at Vapor.io
Links
[00:00:00] Matt: [00:00:00] Hi, this is Matt Trifiro, CMO of edge infrastructure company, vapor IO, and co-chair of the Linux foundation, state of the edge project. Today. I'm here with Nick Barcet senior director of technology strategy at Red Hat. We're going to talk about Nick's background of technology, the future of edge computing and Nick's enviable life of setting edge strategy for red hat while he sails around the world on his boat.
[00:00:22] Hey, Nick, how are you doing today?
[00:00:23] Nick: [00:00:23] I'm doing great, Matt, how are you doing?
[00:00:25]Matt: [00:00:25] I'm doing great. I, you know, the, the most important question is, are you really on a boat and where are you on?
[00:00:31] Nick: [00:00:31] Yes, I'm really on a boat. And if you listen carefully, you might hear a few waves. There's a little bit of wind today. Um, uh, in key West at the moment we've been here for a couple of weeks now, really
[00:00:41] Matt: [00:00:41] enjoyable.
[00:00:42] That's excellent. Do you mostly stay on the coast of the United States or is it really all over the world
[00:00:45] Nick: [00:00:45] so far? We've been, uh, going as far as The Bahamas, but mostly the East coast of the U S. A little bit more convenient
[00:00:53] Matt: [00:00:53] for work. You're truly on the edge, which I think is very appropriate for the show.
[00:00:56] How, how do you get
[00:00:57] Nick: [00:00:57] internet? So I usually say I have a [00:01:00] floating IP. Basically I have four contracts. We use four different providers that gives me unlimited 4g and I load balance between them. That's the best solution I have to prevent, uh, uh, two inches.
[00:01:12] Matt: [00:01:12] You have a technology solution that load balances across four different 4g signals.
[00:01:16] Nick: [00:01:16] Yeah, basically Daisy, Daisy chaining, multiple providers. Yeah. So I'm just, what do force to offer something that works on the moving.
[00:01:25] Matt: [00:01:25] Cool. Yeah. Until Starlink is here, I think it's a great solution. So as you even get started in technology,
[00:01:31] Oh,
[00:01:31] Nick: [00:01:31] I started, uh, when I was eight, I think building a little boards, electronic boards, and then somebody offered me a Tia 57.
[00:01:40] So I. Learn the very basics of proclamation with it got really Intuit bought a TRS 80. Yeah. That dates that old, um, and never left the field. I mean, I I've always had four passions. One of them is computing, sailing, photography, human [00:02:00] studies. In general, I decided that computing was the one that was going to pay for the bills and have fun at the same time.
[00:02:06] Well, it's
[00:02:06] Matt: [00:02:06] nice. And it allows you to allows you to float around the world and your job as a, as a technology strategist, I'm sure allows you to use the human element quite well. So let's talk about edge computing and let's start broadly. So how did you even get involved in edge a
[00:02:19] Nick: [00:02:19] conversation with Matt Hicks who leads?
[00:02:22] Our engineering group told me, Hey, I'm hearing more and more about edge. We have no clue what it is. Can you help us investigate that field? That was. Almost three years ago now. So we, uh, one of my colleague that has now left, shred out, we started, uh, investing, investigating that field a little bit later, provided a little report and we were asked to create what we call an ILT and initiative leadership team within red ads.
[00:02:49] So basically we knew that ed was going to meet our products all across our product branch and we needed to influence the roadmap. Of multiple [00:03:00] products. So we needed a structure that would be across all of our business units to move forward in reasonably synchronized fashion. And that's what we've been doing for the past two
[00:03:10] Matt: [00:03:10] years.
[00:03:11] Yeah. So I think we came into edge around the same time and I remember it being very frothy. I remember, I mean, in fact, one of the most popular questions to ask people was what is your definition of edge, which I don't do anymore, because I think we finally concluded that. Changes, right. I mean, for you, the edge is your boat.
[00:03:29] But I think that one of the things that's happened recently is your definition of edge actually does matter. And so I'm interested in if you've had an evolution of your understanding of edge, and if there's any insight you can offer the audience in terms of how you think about it now, compared to how you may have thought about it at the beginning.
[00:03:45] Nick: [00:03:45] So it's easy to give a definition of edge that covers. All the use cases when you just take a high level stance and say, edge is about putting the processing power closer to where the data is being used or generated. [00:04:00] That's what you'll find on Wikipedia. And I find these definitions absolutely fine. The problem is when you start to want to describe what you're doing at the edge, and here more recently, I've heard from Forester interesting definition, an interesting taxonomy of the edges.
[00:04:17] There is an edge that satisfied the enterprises need. There is an edge that satisfies the customer's needs. There is a, an edge that satisfied the operations of a structure, and there is an edge that he's all about. How do I provide access to. The network. So these four edges seems to be more interesting and allow us to better understand what we are talking about when we talk about a customer use case.
[00:04:46] And sometimes even customer we'll mix two or three of these edges in needs needs. It's not necessarily all about doing one thing, but it can be about doing multiple things that crossover. So, yeah, [00:05:00] edgy is it's a little bit, uh, as ether was saying, uh, 20 years ago, I mean computing or in cloud computing 10
[00:05:09] Matt: [00:05:09] years ago.
[00:05:10] So let's come back to that. Cause I think that's a really interesting element of discussion that I'd like to introduce into this. But before we do that, let's talk a little bit about your role at red hat. What is the director of strategy? What is it that you do at red hat? That's an
[00:05:22] Nick: [00:05:22] interesting question.
[00:05:23] It took me quite a while to figure it out because this position didn't exist. It was created two years ago, as I was transitioning away of my role leading the OpenStack team, basically what I ended up doing years, put people in relation within the virus section of the company, to think together on how to solve a problem that is bigger than any single product.
[00:05:48] I spend time with customers to listen to what they need that we don't have. And I try to distill that information internally to make [00:06:00] sure that we have the same understanding of what, what is the job to be done? What do we need to solve today? And I try to coordinate people in their activity towards solving.
[00:06:11] Matt: [00:06:11] Can you give us an example of an example of one of these engagements that you put together and what it yielded? So
[00:06:17] Nick: [00:06:17] edge is a perfect example. As a question, what is edge? Can you tell us? I had to read, I had to go and meet with a few customers that were tagged as, Oh, these are edge users and talk with them.
[00:06:30] And then I had to restitute that information that I learned. To the management team who then tasked me to try to coordinate our activity around edge internally. What's really interesting in that job from, uh, my perspective is the fact that I have to be listening a lot better than I've ever been a lot better than any single PM.
[00:06:54] Because when we are talking about a new field of interest, What people are [00:07:00] saying is still not yet normalized. And we still have problem with definitions. We still have problems of, are we talking about this layer or that layer? And there, there is a lot of confusion, you know, if I take the average person in the it field and ask them, what is the edge?
[00:07:17] The answer is generally linked to mobile phone and 5g. And while these are important, competent that can constitute edge strategy. They are not the beginning nor the end of it. So I think it's super interesting to be in these early phases and try to bring more clarity in the fog in which we are moving.
[00:07:39] Matt: [00:07:39] Yeah, it can be either be a really exciting or infuriating. It's been an interesting journey for me as well, mostly exciting. Let's see. Seven. I liked sort of sorting through that and identifying things that people have in common that may be using different words for different ways to explain things so that different camps can understand it.
[00:07:53] It's really an interesting role you've created for yourself. So let's go back to the Forester definitions and let's unpack one or two of them. So you mentioned [00:08:00] enterprise, and if I understood the description correctly, there is an edge that. Enterprises will derive value from, or there may be more than one edge.
[00:08:09] So how does that way of looking at it start to provide insights and clarity.
[00:08:14] Nick: [00:08:14] So for a company like red hat, which naturally focuses mostly on the infrastructure, It allows us to look at it from a step higher, which is the application. And when we say enterprise, we are not qualifying anymore. The hardware that is running at the edge, we are talking about what is happening with the data.
[00:08:35] And you have a few patterns in how the data is being used. And when you're looking at the enterprise edge, generally is a data. Flows in and out of the edge towards a central location, which can be a cloud, which can be your private data center, but it's still a very centralized computing, architecture. We as compute resources that are placed nearby the user [00:09:00] to ensure quality of service, to ensure response time, to ensure what's happening.
[00:09:06] If the connection to the internet fails, it's something to improve. On something that could live without the edge and. If you take the second definition, the definition of the operations edge here, you are much more talking about the closed loop. You take data, you process it and you do something with it locally.
[00:09:28] Matt: [00:09:28] What is an operations? What is not an example of a use case and operations? Uh, factually,
[00:09:33] Nick: [00:09:33] okay. You are deploying a new system that will improve quality assurance in the factory and your system is getting to
[00:09:41] Matt: [00:09:41] right. Or I just got my ass kicked by COVID and I need to roboticize my tenure robotics plan is now 3d robotics plan.
[00:09:47] Nick: [00:09:47] And in that case, when you were talking about closed loops, you have fractions of a second to make decisions. And they need to be trustable, but the key point is the production. It's not [00:10:00] how you're going to be reporting to your central location. Central location reporting always exists, but it's not anymore at the core of the
[00:10:08] Matt: [00:10:08] activity.
[00:10:09] Yeah, that's interesting. So you start to identify these patterns, as you mentioned, and I can imagine what the pattern is for the access network piece, which I think was one of the, one of the four. So one of the things you mentioned that surprised me, and I'm interested in you defining it for me, you meant, you mentioned that red hat is an infrastructure company.
[00:10:25] And when I think of infrastructure, I think of like fiber in the ground, I think of like towers and radios and things like that. So explain to me how you think of red hat as an infrastructure company.
[00:10:34] Nick: [00:10:34] Okay. That's interesting because again, that's the same anti clash that we have here. When we say that we mean that we deliver software that help people build infrastructures.
[00:10:44] Generally, we are mentioning servers. We are mentioning networks. We are mentioning connecting applications together, or building storage for us. When we say infrastructure, that's a combination. Of what do you need [00:11:00] to build a, uh, computing infrastructure in terms of software?
[00:11:05] Matt: [00:11:05] Right. But what's really interesting about that.
[00:11:07] As, as you get into, let's say less tolerant edge applications, edge applications that are more demanding on the infrastructure, and we'll just put that in quotes for a second. Um, it would seem to me that. As a software company, you would need to become much more opinionated about the physical infrastructure available, the kind of network you have available, the sort of SLS that can deliver the kind of hardware that you have available, whether it's got a smart NIC or a GPU or something like that.
[00:11:33] Can you talk to me from a software perspective about how you view the physical infrastructure and how you translate an applications demands into software that then eventually has to be implemented on somebody's hardware and how do you get it there and ensure that it's going to do the right thing for you.
[00:11:50] Nick: [00:11:50] So you've described a hammer and the sledge and we are right in the middle. So magically every time an application needs to use a GPU, it needs to be exposed [00:12:00] to it as a, an interface. And that's the role of our software to make these GPU available to the software. So we make the translation from physical to virtual, if you want.
[00:12:12] And we do that for desk. We do that for GPU's. We do that for memory. That's a role of an operating system. And nowadays that's a role of infrastructure, software, such as Cuba that is to present virtualization of those in a way that is even more abstracted so that multiple people can share the same physical heart.
[00:12:32] Matt: [00:12:32] How do you view the relationship between hardware that is on premises and owned by the customer hardware? That's on premises, but maybe owned by AMA well, maybe managed by. Amazon or Microsoft cloud or the equivalent hardware that might be in a nearby data center. I mean, there's a lot of different, you know, whether it's, uh, an outpost box at a Verizon, well, wavelength, you know, an Amazon server at Verizon, it's very different than a Dell [00:13:00] box that I bought that I stuck in a micro data center in the parking lot of my factory.
[00:13:04] How do you modularize all that or don't you, what, what is, what is your view of that?
[00:13:08] Nick: [00:13:08] So our view either that our role is to make those differences almost invisible to the application user, the application developer, and as much as possible to the operator of the edge. So that that's really where our, I read cloud vision, which is now extended to go all the way to the edge provides a tremendous difference.
[00:13:31] So having an open source basis enables us to ensure that. What we're doing is not unique in our customer are not locked in. It's also allows us to collaborate with hundreds of partners, knowing exactly what we're talking about because the code is shared. The, I read part of it is all about allowing the.
[00:13:55] Same interfaces to be presented to the developer while you're deploying [00:14:00] the infrastructure software on vastly different hardware. So sometime it's on a cloud infrastructure, sometimes it's on physical hardware. Sometimes it's on physical hardware that the customer owns. Sometimes it isn't and all of this should be made completely.
[00:14:17] Invisible to those users. When you're talking that we is an API, such as the Kubernetes API, you are generally completely abstracted, and we are now getting used of these attraction in the cloud. When you deploy your humanities environments, use only gain that capability to be. Completely protected from the differences from one cloud to the other, applying it to the edge is our new challenge.
[00:14:42] What we are doing at the moment. It starts by ensuring that we work with multiple hardware vendor with multiple processor architectures, but it also goes into ensuring that. What we have built together. We as either hyperscalers on their platform to enable [00:15:00] our offering, um, AWS or Azure or Google cloud can be reproduced on their wide similar offering, but are.
[00:15:10] Relocated at the customer. And if we can achieve that, then we offer this real ability to be, uh, seeing all of it as an open hybrid cloud. Yeah, that's really
[00:15:26] Matt: [00:15:26] interesting. So do you imagine a world in the future where as an application developer, that's building an edge application, but maybe I'm not at the edge, I'm on a boat, on a building, an edge application, these to run somewhere else on hardware that I.
[00:15:38] Don't ever get to touch and it might be owned by the customer. It might be owned by Amazon, or it might be owned by the telco or so on. Do I supply a manifest or a description of what I need? Like if I have a data sovereignty requirement, I need to be within this geo-fence I need to have this kind of latency.
[00:16:00] [00:15:59] I need to have this kind of resilience. I can tolerate this much jitter. You know, I need this kind of performance from a GPU CPU and so on. And then the infrastructure software figures out where to run that, given the available resources, or, you know, puts out an error that says you can't do it. Or like what's the, what's the long-term vision of this and how close
[00:16:19] Nick: [00:16:19] are we to it?
[00:16:20] Well, I think we are quite far away from it right now. What infrastructure software does in terms of scheduling is pretty good at scheduling within a given locality. And using or exposing the resources available there before we can. Dynamically map a topology of various data center and redirect things to serve the best route.
[00:16:46] I think we will have move on to a new, to a new, to another problem. Or the network will have evolved so much that we don't care that much anymore. About where are things either of these two things will happen before we, we sold it too complex of a [00:17:00] problem to be sold and be financially viable at the end.
[00:17:05] Right now, the problem that we see with our customers are much simpler. It's how do I deploy my new infrastructure in my factory? And what are the minimum requirements? What are the specs for the machine we are going to place there? The difference
[00:17:20] Matt: [00:17:20] between deploying an application on my factory floor and traditional on-prem workload, what makes that edge?
[00:17:27] The main
[00:17:27] Nick: [00:17:27] difference is that generally the location as not being designed to host the computing. So you have all kinds of constraints, depending on where you are, can be space constraints, power constraints, internet access constraint. Sometimes you only have intermittent access to the internet. This is the big difference.
[00:17:47] And you need to be right-sizing this new infrastructure that you're going to be deploying, because if you miss the amount of Ram or the amount of CPU, the cost of deployment [00:18:00] is generally much higher than the cost of the hardware itself. So you don't want to have to redeploy multiple time desserts. It's a little simpler.
[00:18:10] If you are renting an edge location to a provider, because in that case, generally, the provider has more flexibility than you do because the provider is factorized. And therefore, if you need one more server, you can modify the contract. But if you're talking about the on-premise edge owned by the customer, or in other cases where you are renting machines are sent somewhere, that's a crucial element, making sure that your architecture matches the software requirement that you're going to deploy.
[00:18:42] Matt: [00:18:42] Yeah. So it seems that at least in the near term, an application developer, that's building on top of red hat infrastructure software, or anybody else's infrastructure software probably still needs to have some visibility into the hardware it's available. Now you may extract. It's so that the programming models, the same, I [00:19:00] can deliver a Docker image and be confident that's going to run.
[00:19:03] And when it fails, it'll automatically restart and things like that, that these, but I still, if I require a GPU, I have to have confidence that there's going to be a GPU at the other end. And it's probably because I put it there or my customer put it there. Is that, am I hearing that correctly?
[00:19:17] Nick: [00:19:17] Absolutely.
[00:19:18] You can not abstracted, uh, the presence of a GPU. It needs to be there somewhere physically. Now you have strategies. If you already have deployed hardware, you can offload some of, uh, the work that needs to be done. And you can use the regular CPU where GPU would go a lot faster, but maybe the flow of data that you have is not that big.
[00:19:39] But clearly when we are building edge strategy with the customer, we always start from what are the applications that you plans? To deploy before we can even start sizing, what type of hardware is going to be placed there.
[00:19:56] Matt: [00:19:56] Yeah. And in terms of the software that red hat delivers to [00:20:00] your customers, who are your customers and what are they looking for at the edge?
[00:20:05] Nick: [00:20:05] So our customers are across all verticals, right? At works with. All of the top 500 company and much, much more than this, where we see a lot of edge related demand right now are in. The top three where we could name first telco, they are busy deploying 5g, which requires an edge infrastructure, 42 D performance.
[00:20:34] Then automotive, lots of evolution are happening in the auto industry and the car is the. First is competent in the edge for them, but also needs to work on their factories. Yeah, let's see like that. And then the general industrial environment, whether it's all in gas or standard factories are really, really keen on improving their processes [00:21:00] by deploying edge infrastructure.
[00:21:02] But that doesn't stop here. We see opportunities happening in retail, in healthcare, in public sector. And when I say public sectors, that goes from a smart seat to how do I build my next generation of, uh, weapon defense system or RJ go install, uh, the base onto the moon. Really, really varied.
[00:21:24] Matt: [00:21:24] Yeah. When will OpenStack be beyond the moon?
[00:21:25] That's what I want
[00:21:26] Nick: [00:21:26] to know. OpenStack. I don't think at this point, open shifts, very
[00:21:30] Matt: [00:21:30] likely. It's probably been in the, in the atmosphere. But maybe not all the way, man.
[00:21:35] Nick: [00:21:35] Yeah. We've got the communities, uh, OpenShift on ISS that was launched three weeks ago. So that's a very interesting first step. And here it's a typical edge use case.
[00:21:45] Yeah. How do I avoid having to send gigabytes of data back to earth for analysis?
[00:21:51] Matt: [00:21:51] Yeah, actually, that's the one definition I haven't heard. It's the outer shell of my saddle. I got the bumper on my car once when I asked what is the edge of [00:22:00] not the outer shell of my satellite. So let's look at a couple of use cases.
[00:22:03] So let's talk about the telco use case, which I think is really interesting. And I'm seeing the same thing in the market. There's a lot of demand for the virtualization of the network, which requires. Infrastructure software. And a lot of the things that we think of cloud not being centralized, but being out in the field, out in the landscape, what parts of the telco?
[00:22:23] And I don't just wanna say the network. I mean, the, I mean, obviously there's the backend office and the billing and let's, let's, that's traditional enterprise workloads, but in terms of the virtualization of the actual network and maybe the packet core, can you. Talk to me about a deployment that's using red hat software and how that works and what challenges you saw and how you had to adapt OpenShift to accommodate that.
[00:22:42] Nick: [00:22:42] So starting five years ago, we started, uh, working with sail, go on building OpenStack to meet their requirement, to deploy their DNS. And when we did that, we were really focused at the core of the network, because any, in any case, all [00:23:00] the, the end points, uh, an IP connection that you would establish on your phone.
[00:23:04] Would be translated into an IP packet that would go on internet only at a single location. For a provider. So everything was super centralized. So the core was really our focus. And when we did that, we realized that there were a lot of constraints that you needed to put in place to support those VNS seems like enabling all kinds of.
[00:23:27] Quality assurance in how we are going to be delivering resources to a given application. Uh, we are going to segment a computer so that multiple things can happen at the same time without ever disturbing each other. And that was a very interesting, uh, learning. So we, we learned that well, OpenStack, the first thing we started doing in the 5g deployment was.
[00:23:51] Move all this knowledge to humanities, because it was obvious that containerization allowed a lot more [00:24:00] efficiencies than just virtual machines. So we had to reimplement. Everything that we learned on OpenStack. So they would work on OpenShift and that was still at the beginning, focused on the core. And then as you certainly know, when you deploy 5g in order to get lower latency, you need to deploy hardware closer to the antennas and where we only had one level of delegation between the core and the antenna in the 4g model.
[00:24:31] There is now two, even three layers of delegation that you can introduce when you deploy a 5g network. And there suddenly you have new sets of requirements because the difference of your control plane using a five or six CPU's in the core, doesn't make a big deal because a single control plane is going to be controlling tens, hundreds, thousands of servers.
[00:24:58] But when you were talking about an edge [00:25:00] location, a difference of one or two CPU can make a big difference. And you also have requirements such as enabling new hardware, competent like FTGS, or GPU's ensuring that your real-time operating system does deliver the predictability TG supposed to. And that's what we are doing right now with our customer.
[00:25:22] He is finalizing those tax that goes as close as possible to the antenna.
[00:25:27] Matt: [00:25:27] Yeah. You know, it's really interesting. So I got my start, uh, building video games and embedded systems and back in the day, I mean, you know, cause you owned a TRS 80 and all that like discreet timing was how you got anything done.
[00:25:39] You counted cycles to change colors on the screen. And as the world has moved towards the cloud and the internet, you know, there's this really big gap between the cloud and internet developers, where the model is best effort. It works most of the time. Sometimes it'll take longer. Sometimes it won't. So if my Facebook app doesn't reload my [00:26:00] feed for two seconds every two weeks, but most of the time it's fast enough.
[00:26:04] That's fine with me. But when you're running a telco network or an autonomous drone or a factory robot, You don't have that luxury best effort. Doesn't count. Like you have to have discreet, deterministic, reliable. And so can you tell me how those two worlds are meeting, how they're compromising and where the learning exchanges happening?
[00:26:25] There
[00:26:25] Nick: [00:26:25] are two types of way of defining the deterministic events and how you, how are you going to be constrained to source for something? The first one is latency. The time it takes to from, I'm asking you a question to, I get the reply, these latencies, what you see when you're talking about, uh, drone control or, uh, see like that.
[00:26:47] And here, what you have to ensure that that's, you have enough capacity to handle your peak of number of requests. So that's your always below a certain threshold. [00:27:00] You know that your machine can process 50 requests per millisecond, weld meat, you do simple math and you add enough machine to do, uh, up to 500.
[00:27:10] If that's your capacity planning for a given location, that's fairly easy to do because system like this is just a matter of rightsizing it. However, when it becomes really complex from our perspective, Where do we translate hardware to virtual ease? Every time we have to do analog to digital transformation, example of this is transforming a radio signal into a string of bits on my computer, or I am watching for some event and I need to make sure that I'm here to listen when you don't cures.
[00:27:49] And I need to do it at the right frequency. In fact, it's, every time we have to fight with the four years low that we have to engineer [00:28:00] computers that are super predictable on when they are doing something. And people think that when we say real-time computing, We mean real fast computing? No, actually we slow things down to ensure the things happen at the right beat.
[00:28:15] Matt: [00:28:15] Terministic is probably a better word, but very few people understand what that means, but yes, I definitely know what you mean. Predictable, reliable, consistent is what you need. Exactly. Whereas the internet is like, let's just, you know, the bits can arrive in any order and hopefully it works.
[00:28:30] Nick: [00:28:30] That's it. So every time you do these conversion from analog to digital.
[00:28:35] Either you can do a sampling that is so fast that you don't care about missing a beat, and you have to be many times faster than what's happening in real life. Or you need to be on beat. It's like in a group, if you are not playing on the same beat as the rest of the group, you're going to. Mess it up and we can do that.
[00:28:54] So the tolerance there are really, really key and [00:29:00] fine tuning software for a particular piece of hardware is the difficulty. Every time we have to deal with this.
[00:29:06] Matt: [00:29:06] That's really interesting. That's really interesting. So what about resilience? I mean, because servers fail, fiber gets cut radios, go down, power goes out, backup batteries drain.
[00:29:19] Is there anything different about providing resilience in a deterministic system that you've had to do from a red hat perspective?
[00:29:24] Nick: [00:29:24] Not that much resilience is always the same and becomes more problematic as you add to your requirements. If you have a system that needs to be 99% of the time available, that's really easy to set up.
[00:29:40] If you need something that is available at nine, 9.9995 nines, that becomes really problematic because we are talking about. Ours have, uh, just a few ours, uh, outage during a full year. And it's really complex to design such a system. It requires [00:30:00] understanding not only what's happening within a given application, but within the entire environment, in order to be able to.
[00:30:09] Predict and correct as soon as possible, even before they arrive in order to meet these types of requirements. So that's where the difficulty lies as competing. Hasn't changed anything to this, except that you have more constraint, you have less space, you have less power. So sometimes when people tell me I only have room for one server and I want that to be super available, five nights, there is no solution for this.
[00:30:36] Matt: [00:30:36] Yeah. You know, you know what I thought you were going to say, this is interesting. And, and maybe you, you, this is implied. Cause you think this world coming from the software world, but what I thought you might say is we're moving from a world of. Fault tolerance through mechanical redundancy to a world of high availability through software, because I'm certainly seeing that trend.
[00:30:56] And obviously with systems like Kubernetes, like high [00:31:00] availability is built in it's part of it. And absolutely. Yeah. So let me talk about that a little bit.
[00:31:05] Nick: [00:31:05] Absolutely. But that's true when you're talking at the cloud computing level where the number of computer years, uh, and greater than three. Well equal or greater than three, then you can start applying all kinds of software to make that completely transparent to the software developer or very close to transport is a software developer and, and, and I availabilities for software.
[00:31:26] But when you are talking about edge computing, where when people can deploy three server, You're super happy. I mean, it's like you have a party because it's not every day, but three server can be hosted, uh, in every store, uh, of a given company sometime too is going to be there maximum. And that's where you can do everything you want with your software.
[00:31:49] If you don't have a minimum of three server, we don't know how to provide very high vulnerability. There are physical limits to what software can do.
[00:31:58] Matt: [00:31:58] Yeah, that's a really, that's an interesting [00:32:00] distinction on that spectrum of, you know, cord edge, which is at the point where you've got a system where you can implement high availability for resilience.
[00:32:07] And at some point, yes, it's just not practical. You've got one CPU, one server on the windmill or, or whatever it is. And you just have to make that as reliable as you can possibly make
[00:32:17] Nick: [00:32:17] it. So that means allowing for all the strategies that you have here, like for, uh, 5g networks. What the operators do is try to never rely on a single antenna to serve a given location so that they have overlapping between their different antennas.
[00:32:37] And if one fails, the other one has a little bit of extra capacity to take over the traffic that was handled with the antennas that failed. There is other strategies like, okay, I can not do any mean of high availability on a given site. However, If this site disappears for a reason or another, I'd rather make sure I know about it state so that I can [00:33:00] replace it with another.
[00:33:01] So you implement disaster recovery scenario between multiple sites or you design your software to be completely stainless. Which allows for a lot more flexibility in switching from one side to another, but you have to think about it because suddenly your physical environment is coming with lots of constraints and you don't want the software, meaning the application to have to worry about it, but you want the infrastructure to enable, uh, for the solution.
[00:33:33] And that's based on, um, Intelligent and creative thinking at the time where you're designing this edge infrastructure.
[00:33:42] Matt: [00:33:42] Yeah. Some really interesting innovation is going to happen there that thank you for pointing that out. Let's switch gears a little bit. So when I think of red hat, I think of the gold standard of.
[00:33:51] Commercialized open source software. And you've said the future of edge computing will absolutely be open. Can you tell me what that [00:34:00] means? What the role of open and open software will mean an edge computing and what you're seeing and where there are opportunities or where there are some danger signs?
[00:34:09] Nick: [00:34:09] I believe I touched on that, uh, when I, uh, gave a definition of open hybrid cloud, including the edge, but. Let me try to refine that a bit. So for me, the problem for a given customer, IE, that there is a big risk that the choices you make early on on these ed strategy will add to an already very big set of constraints.
[00:34:33] Because if I choose to go with a given vendor that has a super vertical integration, From the cloud to the edge, the day I add another edge use case, which is not covered appropriately with that vendor. I am at risk to have to redeploy everything in another way. For me, the open ness that we have. [00:35:00] Enables two things.
[00:35:02] The first thing is ensuring that the same infrastructure software works the same way across multiple vendors. So that you can easily add another vendor. Let, let me take a more concrete example. I have decided to deploy my, uh, my solution using hosted edge equipment. And for my first, for example, the vendor that has multiple small data center all across the country.
[00:35:31] And based on my analysis, all let's say I'm a retail company. All of my store are close enough to one of those locations. So that solves my problem completely. And I'm going to work with this company because I don't have any other needs next year. Suddenly my company decides to open a series of store in Europe, and I realized that my edge vendor.
[00:35:56] That was available all over the place in the U S [00:36:00] he's not available in Europe. What do I do? Do I start from scratch or do I have something that abstract me from the vendor I've chosen and enables me to deploy elsewhere? So that's the first part. The second part is even we as a vendor that has availability across software vendor, that availability across multiple hardware vendor.
[00:36:25] There are risks that we decide to discontinue a given product line and put people in the same amount of trouble. However, when you're dealing with open-source software, there is never a case where it was single company. Offers that piece of software, there are generally multiple options and that's another way of protecting you from having to redo everything from scratch.
[00:36:49] So that's why I think that the edge cannot be free. 12 can not bring maximum benefits without being open, because if you are [00:37:00] not protected against these two risks, You're going to have a very, very costly edge.
[00:37:05] Matt: [00:37:05] Yeah. I, I, I understand that. I understand that. What about the aspect of open source software in particular where the idea is that an industry, a new industry, like edge computing can accelerate itself by having the stakeholders agree on a common substrate?
[00:37:24] That they all contribute to, to advance its philosophies. Maybe Kubernetes is a good example of this, where it would be ridiculous for someone to create a new container orchestration system. At this point, when you've got one that's, you know, the top engineers in the world have been banging on and will continue being on refining.
[00:37:41] So do you see other projects? And if you can name specific projects, I'd be interested just. That you have a particular interest in that you think are because of this kind of open source collaborations are accelerating or advancing that common substrate for edge computing or not? Well,
[00:37:55] Nick: [00:37:55] the projects regarding building a CIC pipeline [00:38:00] get ops operation automation, whether it's in the network or into software.
[00:38:04] All of these competencies are absolutely fundamental to building these your edge. And, you know, is a small part of that problem. And there are areas where we see needs, where we don't see yet a project matching it and where we are investing in the research necessary to solve it. Like how do I deal with a multiple vendor hardware life cycle?
[00:38:28] There are plenty of solution to control a specific type of hardware. But when you start missing out we're vendor managing the hardware life cycle is a nightmare. Even OpenShift. We often say OpenShift is Kubernetes. That's true. But OpenShift is a lot more than Kubernetes is it's Kubernetes plus service mesh, plus CICT towards plus many others.
[00:38:51] So I think that the question is how do you select the right projects so that the [00:39:00] are not maintained by a single vendor? And how do you ensure that you have a capacity to maintain those projects over time? Whether it's by contracting with someone else like red hat or by ensuring that you have the team to do so for yourself.
[00:39:18] And that's what open source is about. It's making sure you're not doing things by yourself because we can go so much faster. And we've proven that now multiple times then, uh, regular software vendors, that it would be. It's stupid to be approaching the problem any other way.
[00:39:37] Matt: [00:39:37] It's a really interesting time to be alive from a software perspective.
[00:39:41] I completely agree. I mean, I grew up in the world where the last thing you thought of doing was letting someone look at your source code. And now it's kind of the first thing it's like, which parts are we gonna open source? It's really interesting. So let's, let's just wrap up a little, um, what do you see as the biggest near-term challenges facing edge computing?
[00:39:58] Nick: [00:39:58] There are so many [00:40:00] I can say right now, the biggest challenge. Yes for the hardware vendor and the software vendor to start normalizing what they're doing. Because right now, when we're talking about the single board computers, you have has many boot systems that there are vendors to deal with these environments.
[00:40:21] Um, you have as many message cues as you have vendors. You have. So trying to start normalizing this, this has already started, but it's going to take quite a while for it to reach every vertical industry. Is any
[00:40:40] Matt: [00:40:40] of that normalization organized? Is there any group doing, doing that
[00:40:44] Nick: [00:40:44] multiples? And generally they are, um, uh, by verticals and there is now some, uh, effort that are cross-pollinating across industries, but it's still quite interesting that.
[00:40:57] Even two years into it, I [00:41:00] can get into a customer and hear about a new way to, for message to be sent from ATB. I don't know if you have the same experience, but, um, it seemed that we've been really, really good at making multiple VHS and multiple better max things with spaces before the computing world started looking into it and say, Hey, that's.
[00:41:22] Weird. We need to simplify
[00:41:23] Matt: [00:41:23] that. Yeah. It seems like innovation and normalization are a high tension, have a high tension relationship and figuring out how to align. Those is really the challenge. Yeah. Yeah. That's really interesting. So let's, let's look out into the future. Let's look, you know, 18 to 24 months, where do you think we're going to be?
[00:41:39] What's gonna be the most interesting thing happening in edge. So
[00:41:42] Nick: [00:41:42] on one side you have the evolution of networking. We were talking about it regarding my boat, but this evolution that is being brought by 5g deployments by Starlink or multiple, uh, space-based uh, networking is going to change a [00:42:00] lot of things on how we approach the networking problem.
[00:42:04] Know, I was telling you, I have multiple providers using 4g today to ensure high availability. The high availability of network is going to become a problem of the past because of this. And it's, if we look at 20 years from now, we even have hoped to have instantaneous communication at any point of space, uh, through, uh, Uh, as in quantum, we could hope that 20 years from now, we could have instantaneous communication from any
[00:42:34] Matt: [00:42:34] point instantaneous as an, as in quantum or instantaneous as a
[00:42:38] Nick: [00:42:38] spiel.
[00:42:39] Oh, that's in the, in the timeframe that you were looking at, you're looking at just 5g and satellite based communication. Are going to revolutionize our way to design because we'll have so many offerings that will essentially become cheap enough that we'll be able to have two or three, regardless of the use case.
[00:42:59] So that's something [00:43:00] that really looking forward to not only because I live on a boat, but because of the requirements I see, uh, my customer, the other, uh, evolution that is really interesting is how we are using. Processing power in. More specialized units,
[00:43:20] Matt: [00:43:20] like smart necks and SPGs and things like that.
[00:43:22] Yeah. Okay. That's,
[00:43:23] Nick: [00:43:23] that's what I'm talking about. The fact that you have DPU is available in smart mix, and we can use those to do things in parallel of what the main CPU is doing. And I believe that this specialization of resources is going to enable a new generation of software. That provides for solving problem quicker and much cheaper than we were ever before.
[00:43:48] We've seen that revolution happen through GPU's, but we are only at the beginning of this revolution. And it's going to go through all kinds of things and all kinds of [00:44:00] discoveries. So I don't know, where is the next major breakthrough you mentioned
[00:44:04] Matt: [00:44:04] earlier in the interview that one of the biggest challenges you're confronting is that gap between analog to digital and wondering if the specialized processors smart NICs and FPJ and things like that are starting a potential path to solving a great deal of that challenge.
[00:44:19] Do you see that.
[00:44:20] Nick: [00:44:20] I don't know if it's going to solve it completely, but I know we've already built a proof of concept with Nvidia where we're using GPU to do that conversion. Hmm, that's already starting to happen now, does it prevent us from requiring a real-time current loans in order to deal with that signal?
[00:44:37] I'm not sure yet. I've not been involved too close. Excellent question. And I'll talk with it with my colleague Franklin, build that together.
[00:44:45] Matt: [00:44:45] Yeah, it's really interesting. Thank you, Nick. Thank you for joining us from key West today. And, uh, really appreciate you joining the show and sharing your insight with our audience.
[00:44:56] Um, if people want to get ahold of you online, how can they find you
[00:44:59] Nick: [00:44:59] quite easy? I [00:45:00] am named nijaba on the, almost a old social meetup platform. NIJABA, which stands for Nicholas James Barcet, which is my birth name.
[00:45:08] Matt: [00:45:08] Excellent. Well, again, thank you very much. And, uh, I'm excited to see, uh, more red hat innovations in this
[00:45:13] Nick: [00:45:13] space.
[00:45:13] Thank you very much. It's really a pleasure talking with you today.