 Hey, welcome everyone. Jeff Frick here with theCUBE. We're on the ground in San Ramon, California at the GE Digital Compound. It's growing, it's like 1,200 people, 1,500 people. I can't even keep track every time we come out. It grows by another 500 people. So kudos to the team for finding so many great software engineers in a crazy environment. We're excited to be joined by our next guest, Mark Thomas Schmidt. And you are the chief architect of Predix for GE Digital. Welcome. Thank you, thank you very much. So that's not an insignificant role. So cloud is the hottest thing going on. GE is all about industrial internet. You guys come up with the industrial internet version of the cloud, which is Predix. Tell us about it. Yeah, it's a really interesting challenge. And one doesn't often get the chance that I get here to actually build a platform from scratch. So that's what we've done here in San Ramon over the past couple of years. Basically, we've been given the opportunity to build an entirely new platform. Platform that is very focused on industrial internet applications, which is different from most of the other cloud platforms that we have out there. Poses some interesting challenges for me as an architect. How do you build this out? One of the fundamental decisions we made early on is that we wanted to innovate in the industrial internet space and we want to use open source as much as possible. So the general platform is an open source-based platform. And we're implementing a layer on top of that that maniacally focuses on the use cases that we see in the industrial internet. You mentioned that the cloud-based platform is very important. Obviously, we need the scale of the cloud-based resources out there. But another characteristic of predicts that differentiates it from some of the more generic cloud offerings that you have out there, it's a hybrid platform. It's a platform that basically has one foot in the real world where the assets are. We need to have a presence there and the other foot in the cloud. And making the most out of that combination is the most interesting challenge I see in this one. Yeah, that brings up a great point because we are all the time, right? Lights just too slow. And so the original cloud AWS was really test-depth, not really mission-critical. I throw some stuff up, I scan my card, I'm on the cloud. But in an industrial internet application, in a production application with jets and locomotives and power supplies, these are big, heavy things that cost a lot of money and move a lot of stuff. So how do you deal with the latency? And how did you get to, how do you separate what's on the edge and what's in the cloud? Because the cloud brings a lot of benefits to big data. But moving the data can be tough. Great question. And I think you nicely put the constraints that we have forward. The constraints that we have to speed of light, it basically means that there is a certain latency, inherently not only the speed of light, also the latency in the network itself. There's a certain latency that you have between the point where an asset out there generates data, you have a chance to analyze it, and then you want to react to it. So that motivates some of the work that we're doing to push workloads out of the cloud closer to the asset. Those workloads that actually affect the asset, that immediately kind of analyze the data and in very short time need to take action based on that. So those are typically the workloads that we help our customers to move closer to the assets. The cloud gives us the elbow room, if we will, it gives us the scale that we need to actually build out analytics. It basically is the only place that we have in this ecosystem to aggregate all the data, not only the data generated by one asset or a set of assets, but all the assets in the fleet. You need the scale of the cloud to basically put them on a big pile, you mentioned big data, but basically a big data analytics platform, so you need the scale of the cloud to actually do that big data analytics. But what's special about our platform is we then basically need the results, the insights, the models that we generate with the power of the cloud. We need to be able to push them back to where they can be made operational, and that often is closer to the devices. As you said, the topology out there is for industrial internet is much more complicated than what you see, for example, in the consumer IoT. Consumer IoT, I have a smart thermostat in my house, right? And next to the cloud, I stream some data. You don't need much intermediate hopes. You don't need to worry too much about the speed of light in those scenarios. For the industrial assets that we have is you usually have controllers very close to an individual asset. You often have, over a system of assets, you have gateways, you have aggregation points that basically deal with it, and then you hit the cloud. And the approach that we're taking is we're making use of that entire spectrum of compute nodes from the big data centers where you aggregate them all together to the more distributed ones where you have gateways out there in the real world, down to the sensors that basically still have some compute power. You can actually use them. You can do some smarts on them. So our challenge is to weave all those compute nodes into a homogenous kind of system that our customers, then, they shouldn't have to worry about do-or-do-it-is in a cloud, do-or-do-it-is at the edge. What we want them to do is we want them to focus on what's the end-to-end business application that you want to implement. Focus on writing your analytics. Don't worry about, we'll take care of that. That's the kind of fabric that we're implementing, making it possible to move applications, move workloads from cloud to the edge, and the other way around. You didn't want to try anything hard, huh? I mean, and the other thing too, I mean, we've all been to data centers, right? And they're very homogenous, and everything's very controlled, and that's not what's happening out on the edge, right? You've got all kinds of challenges for location, temperature, vibration, power. How are you kind of addressing that? And then as you mentioned, the edge isn't even really edge. There's sensors, there's gateways, there's actuators, there's things that make things happen. How do you kind of split all these things up and decide what goes where? And manage them. That's part of the secret source comment, where we need to help our customers. See, many of our customers are extremely savvy in terms of understanding their OT infrastructure. Clearly, we also have people that understand their IT infrastructure, bringing those together. And basically, in many cases, what we do is we actually, for the first time in an enterprise, we make those two groups talk to each other. We make the OT people actually talk to the IT people. And the way we do this is by giving them options, where basically, this is what the platform does. The platform basically should, a good platform, should make it so that you, as a consumer, don't have to worry about all the nitty-gritties under the cover. So our mission is to kind of introduce that level of abstraction. We have a component in the platform that we call the Edge Manager, for example. That's a component that knows about all the devices out there, is capable of establishing secure, security is extremely important for us, especially when it comes to pushing stuff closer to the assets, a secure communication with those assets, and then managing the pushing of data from the edge to the cloud, and the other way around applications from the cloud to the data that are out there. So this is basically our challenge from a platform perspective, implement that fabric, implement that mobility, so to speak, of data in one way and applications the other way around. Okay, I hope you don't have much going on this afternoon, because we might not wrap here for a while. The question, open source. You said open source is a really important piece. A, why? And B, was that tough to get into kind of the broader ecosystem here at GE of the benefits of the open source approach? I have to say, and let me answer the second part, to my surprise, a little bit, GE being a traditional company, very easy. I think the people here in the center in GE Digital quickly got that, unless we focus, see, we cannot catch up to where Google and Facebook and the likes have come over the past 10 years and basically building out all the infrastructure that one needs to do big data analytics. And there's no reason for us to reinvent this. So I think the easy sell was we wanted to go with open source where possible. Whenever somebody comes to our product management with a new requirement, they basically say, this is a feature that you should add to the platform. The first thing that my team does is they go out and say, is there an open source component that could do that? If not, then we innovate on top of that. So in a way, we're standing on the shoulders of open source giants. I mean, the last generation, if you will, of platforms that have been built around our system of engagement, where we've basically taken big data that people produce and relate them will benefit. We couldn't do what we're currently doing without those pioneers having done that. Now, what we're doing is we're taking that and we're applying it to a very different use case. So it's not people that produce the data, it's wind turbines. They tweet all the time. They tweet a lot more than my 16-year-old daughter does. And most of the wind turbine tweets are extremely boring. So it's similar kind of data that we deal with, but very different quality. So we benefit from technology that has been generated, but we need to adapt it. We need to kind of give it a tweak. So the picture that we often draw, when we draw architecture pictures, I sometimes jokingly say, there's 90% of our platform is open source and we really write that 10% on top, but that's the differentiator. That's where we apply the expertise that the GE business has over, what do you actually do with wind turbine data? How do you analyze data from, how do you make predictions of the behavior of machines that hardly ever fail? It's very difficult to kind of find the needle in that big data haystack if you're dealing with very reliable machinery, for example. So applying basically generic technologies, open source, and putting the secret source that only GE and a few other companies that understand the industry can actually put on top of that, that's our mission. The open source thing is fascinating. We covered the Open Compute Project Summit, which is basically Facebook's hardware spec that they've opened source to everybody because they want to share the wealth and they don't consider a competitive differentiator. It's really interesting times, but you guys are not into tweets and pretty pictures from the weekend, you're into big, heavy industrial machines. And so how do you get people kind of on ramped into the platform? What are the apps that help them get in and then integrate into what is already, as you said, a big existing system that seems to be working pretty well? Very important point. We're building a platform, we're building a platform we ground up because we see there is a tremendous spectrum of potential applications out there, but to your point, I think, to drive adoption of the platform, we need to illustrate to people what you can do with that. So as you probably know, in parallel, almost in parallel to us building the platform of predicts, we also built the first application, the asset performance management system, APM, that basically takes the predicts platform and applies it to a very popular set of use cases to illustrate to people, this is what you could do with the platform. We'll keep doing that. We'll keep doing it ourselves in GE and we're building out an ecosystem of partners that basically specialize sometimes in particular verticals, sometimes in more cross-cutting functions. So the ecosystem, I think, will help us to illustrate to people this is what you could do with a platform. My job is to build a platform so that all those different players, be they very small, be they very big, that they can translate their expertise quickly into industrial applications. But absolutely, we need those solutions. We need to inspire people, basically show them what's possible. What I'm envisioning, what happens, is that the first wave of adoption around predicts is probably twofold. There are those that adopt predicts through the solutions that we built on top of this, APM being the prime example. And then sometimes they come to a point where they say, okay, I got the out-of-the-box capabilities that GE, APM delivers to me, but I want to extend it a little bit. So those are basically platform users. They use the platform as an extension vehicle to those solutions. And then there's the others that basically take a bolder stance and they basically say, I want to build this myself. Give me a platform that makes it very easy for me. I have this idea of the dashboard that I want to implement on top of my wind turbine fleet across the world. My job is to give them a platform that within a few months actually gets them to a point where they can have an operational system. So we need to understand, we can't just give them a cloud platform and say knock yourself out. Here's a database, here's a messaging system. You know what to do because they probably know what to do but it takes them way too long. So what we need to do is we need to understand the specific patterns. What are the applications that people typically build? And then we harden those patterns and we harden them in the form of services, microservices that run on a platform. And there's a difference between, in some instances there's the difference between the applications you run at the edge and the applications you run in the cloud. Edge applications, by the nature of the beast, they tend to be more focused. They tend to be, we have a smaller footprint where you can run them. But they're also, in the grand scheme of things, they're more the kind of applications that help you to sort through the data, to filter the data, to quickly analyze potential problems in the data you see and take action. While the application that they deploy into the cloud, they're more the kind of, let's take a look at big pilot data. Let's kind of just do some deep learning over this. Let's spot some insights that we couldn't, that people looking at the data couldn't see themselves. So what we're doing is we're building out this platform that gives you the same capabilities at the edge and in the cloud, but then some extra in the cloud that supports those particular patterns that we see in the cloud, some extra at the edge that support those things. So normally detection at the edge is more important. Deep learning in the cloud is more important. Right, so you're so, it's really kind of workload specific courses for courses, kind of what gets done on the edge versus what comes in. So the other kind of concept I think is very interesting to what you guys are onto is historically we look back, you know, we see what happened, we have reports, now we're trying to do more predictive analytics as to what's happening and then even more prescriptive analytics, right? Get ahead of the curve. But where it gets much more interesting is when you start to look at the problem through a completely different lens and to be able to really look at your business in a different way, that's not just whether this piece of machine is going to fail or not or win. And are you seeing some of that kind of next order benefits coming out of the use of this type of machine? I do, and I think I do it. Again, you described it very nicely. This is the phases that most people go through. It's, in many cases, actually there's a lot of benefit. I don't want to belittle it. There's a lot of benefit in just getting a handle on how are my machines out there doing it. Put it in a dashboard and alarm people if something goes wrong, that in itself. It has no deep analytics in there, but that in itself tremendous value because it gives you the ability to much more quickly react. And then as you said, what you really want to do is you want to take that pile of data and instead of just alarming a person, then you take action, you want to predict. You want to say there might be something wrong, you want to do something about it. So this is where everybody that I see at the moment kind of starts. The ones that have been on that journey for a little while, they take a step back and they say, predictive maintenance is interesting, but what else could we do? Could we actually maybe, for example, use our footprint in a particular industry, say in aviation, could we use this to go into adjacent industries? Could we, we're in the business of basically flying airplanes, but could we help the airports, for example, to implement a more efficient end-to-end system? So I think we're getting to the point where people realize once they get a good handle on their digitalization themselves, they might tap into surrounding areas. And as the entire industry becomes more digitized, I think the synergies will be very important. And whoever I think jumps in there first and says, I'm not only going to focus on what I'm currently doing, I actually want to benefit from everybody digitizing their stuff. And I want to run cross-cutting workforce. I basically want to mix and match some of those patterns out there. That would be very important. And what about kind of the ecosystem, whether it's existing infrastructure that people already have in place, maybe it's analog, they haven't digitized everything, and they're not going to rip and replace everything, but they want to start taking advantage of some of these things that you can do with Pridic's cloud. How do you kind of work with the ecosystem, both the ecosystem of kind of the installed base of industrial goods that are there and running and working? As she said, everything works pretty well now. Versus kind of the new stuff and the new software partners and the new kind of greenfield applications where people are getting excited about cloud and McDade. We see both. For obvious reasons, we see a lot more of the ones whether it's kind of brown fields where you have an existing installation. Some of them are instrumented in an excellent way. The more valuable the assets, the better the instrumentation. But not the random example is, we just recently closed a really nice deal with Shinta, the elevator company. And their elevators are not highly instrumented. There was no reason in the past to do that. So what we're basically working with them and we're similar ones, how can you, after the fact, basically instrument some of those assets? You need to come up with clever solutions, price plays a role. How can you basically come up with simple instrumentation solutions that help them to kind of turn an existing, relatively old fashioned situation into something that you can digitize and then take. So I think there's tremendous opportunity in instrumenting those brown fields. I have to admit, I find the greenfields more interesting because you can do more there. And we see a lot, for example, the GE Current folks. They have some really interesting things when it comes to building automation. You know, basically scenarios where you have many, many little things. And the interesting thing where we can help with is not so much optimizing the use of an individual one, a gas turbine, it's absolutely valuable to take the 200 cents that it has and optimize the heck out of it. This is more about many, many things that by themselves are not that interesting, but in combination are very interesting. So that's kind of a trend that I see this kind of system of assets optimization is very interesting. And I see this more in the greenfields and the brown fields. So tell the elevator guys, we did an interview one time and basically elevators are a really good predictor of health of the tenants. And these guys had like hundreds of floors of buildings, I think in downtown Tokyo, and they were analyzing the elevator data and they could tell whose company was in trouble and jump on it. So there's all kind of second order value. All right, Mark, well, so you're sitting in the cappered seat, you're kind of driving this bus in terms of the architecture, in terms of challenges, next kind of challenges that you're trying to overcome, what are some of those things and then what are some of the things you're really excited about, six months down the road, 12 months down the road and I can't tell us any secret stuff unless you really want to tell some secret stuff. But what are you excited about? What's getting you up in the morning? What's getting me up, it's really two things. And one of them we've focused on here. I think there's tremendous potential in this edge play that we're playing out there. I think we have some pretty cool stuff out there already, but there's more that we can do. I think what we've laid out is we have the fundamental infrastructure for us to actually manage edge-to-cloud workloads. Now the next step in this one is how can we add patterns on top of that? How can we basically spot, we're enabling people to do edge-to-cloud, now we need to be smart on seeing what do they typically do and how can we make it a lot easier for them to do that and I think there's a lot that we can do around specific types of analytics that we see people using over and over. What we give them today is we give them the ability to run some analytics, right? I think we want to be smarter about what particular do they do. I think in the edge space, also a very interesting topic for us to explore is how do we benefit from having networks of edge devices and in most kind of traditional installations, edge topologies that moment are hierarchically. You have sensors, you have controllers, you have gateways, you have the cloud, right? What we see happening more and more is that at the higher end, at the gateway level, for example, at the controller level, some of our customers start thinking about having them communicate between each other, so rather than always just going to the mothership to cloud, maybe they can amongst themselves do something interesting. So that's one of the trends that we're currently seeing and this is all kind of very much connected to the other trend, which is the digital twin. It's a concept that we've been supporting for a while, we encourage our customers to think about what they're doing not so much as application development but twin development, right? So we basically want them to think about this as if you have assets out there, don't just think about putting a dashboard on top of them. Basically think about creating a twin for each individual asset you have out there. Build systems of those twins, understand what information you can extract from those twins. To build those twins, what you basically do is you take any information that you get from the devices and then you mix it up with other information you have. You may have some card drawings of the asset. You may have some simulation models. You may have some physics model that describes their behavior. You may have the maintenance report that somebody wrote when they last climbed up the wind drone, right? So bring all those things together and then make them operational. That's our challenge and I think we've taken good steps to do that in our asset service and things around that. I think the challenge that we're facing is now, how can we take that twin and push more and more of it out there to the edge? We basically want to think about this as a twin fabric and operating system, if you will, for twins. Where you have the model of the twin, you generate that model, you model that in the cloud and then you take pieces of it and you say, you know, that information about that individual wind turbine, the controller on that turbine should have that. That information about the wind farm and how it should optimize its use, the gateway on that wind farm should have it. So basically pushing the twin out in the edges is another interesting thing they were doing. As you're talking, I'm thinking, did you solve the problem of light physics problem? Because what you're talking about really is where speed and latency is an issue, you move the compute to the edge. When it's not an issue, you can move it back into the cloud. So really it is kind of this hybrid approach based on optimizing for the process is really what you guys are all about. That's a great way to put it. We're faster than the speed of light and the trick we play is we just move the gold post, that's exactly right. I love it, all right. Well Mark, thanks for taking a few minutes out of your busy day and congratulations and we look forward to continued updates as you kind of move down this journey. Thank you very much. Absolutely. I'm Jeff Frick, you're watching theCUBE. We'll see you next time. Thanks for watching.