 Live from San Francisco, celebrating 10 years of high-tech coverage, it's theCUBE. Covering VMworld 2019. Brought to you by VMware and its ecosystem partners. Welcome back to theCUBE, it's VMworld 2019, our 10th year, wall-to-wall coverage, three days, two sets, lots of content. I'm Stu Miniman, my co-host is Justin Warren and one of the big stories coming into the show is VMware actually went on an acquisition spree. A whole number of acquisitions, Boston-based, carbon black, over $2 billion, Pivotal brought back into the fold for also, you know, around that ballpark of money. I'm happy to welcome to the program one of those acquisitions. Sachin Kati is sitting to my right, he's the co-founder of Oujana, he's also a professor at Stanford University. Thank you so much for joining us. Pleasure to be here. And joining us also for the segment, Shaker Iyer, the executive vice president and general manager of Telco Edge Cloud at VMware. Shaker said, yes, there's a lot of acquisitions, not to play favorites, but maybe this is his favorite. Shaker, thanks for coming back. No question about it. All right, so Sachin, you know, boy, you know, the Palo Alto, you know, Stanford connection, we were thinking back, you know, the founders of VMware, of course, you know, came from Stanford, many acquisitions over the year, including the mega NYSERA acquisition, you know, quite a few years ago, I came out of Stanford. Give us, what was the genesis and the why of Oujana? Yeah, it's an actually interesting Stanford connection too. So I've been a faculty at Stanford for the last 10 years and I've seen the SDN movement very close and up front. And one of the dirty secrets of SDN is it makes the networks programmable, but someone still has to write the programs. And so that's usually a very complex task. And the thesis behind the company was, can we use AI to learn how to program the network rather than humans having to program the network to do management or optimization? So the vision really was, can we build a network that learns how to optimize itself, learns how to manage itself? And the technology we are building is this AI pipeline that basically tries to deliver on that for mobile networks. Yeah, that's great, Sachin. You know, my background is networking and it feels like forever we've been hugging. Well, we need to get people from the CLI over to the GUI, but we know in today's really complex world, whether it's AI or just automation, humans will not be able to keep up with it. And we know that that's where a lot of the errors would happen is when we insert humans into doing some of this. So what are some of the key drivers that make this solution possible today that it might not have been able to do done when Martin was first rolling out the first SDN? Yeah, I'll talk about it in three dimensions. The one is, why do we need it today, right? And then what is happening that is enabling this today, right? So apart from what I talked about SDN, I think the other big driver is, the way I like to think about it is that the internet is going from a means of consumption to a means of control and interaction, right? So increasingly the applications we see driving the next big decade are where we are controlling things remotely over the network, like a self-driving car, or we are interacting with very highly rich visual content like AR and VR. So the applications are becoming a lot more demanding on the network. At the same time, the network is going through a phase of opening up and becoming disaggregated. So network complexity is increasing significantly. So the motivation behind the company and why I thought that it was the right time to start the company was these two trends are going to collide with 5G coming along. The applications that are driving 5G and the network complexity increasing with 5G. So that's why we started the company. What actually is enabling this is the fact that we have seen a lot of progress with AI over the last few years. It hadn't really been applied at scale to networks and specifically mobile networks, so we definitely saw an opportunity there. But increasingly a lot of the infrastructure that was being deployed, there was more and more telemetry available. There was more and more data becoming available. And that also obviously feeds this whole engine. So I think the availability of all of these big data technologies, more data coming in from the network and the need because of these applications and network complexity, I think it is a perfect confluence. There's lots of AI floating around at the moment and there's different flavors of it as well. So there's machine learning, there's AI. So when you say that there's AI behind this, what particular kind of machine learning or AI are you using to drive these networks? There's a few different techniques because the problems we solve are a normally detection of when problems are happening in the network, predicting how network conditions are going to evolve. For example, predicting what your device's throughput is going to be in the next 30 seconds. For example, we are also learning how to control the knobs in the network using AI techniques. So each of these has different classes of AI techniques. So for example, for control, we are using reinforcement learning, which is the same technique that Google used to kind of win on AlphaGo. So how do you learn how to play a game basically? But here the game you're playing is optimizing the network. But for the others, it's recurrent neural networks to do predictions on time series data. So I think it's a combination of techniques. I wouldn't get too worried with the techniques. It's ultimately about what is the problem you're trying to solve, and then we pick the right technique to solve it. And so on that, because AI is actually kind of stupid in that it doesn't know what an optimized network looks like. We have to show it what that is. So how do you actually train these systems to understand what is an optimized network? How does a telco define this is what my network optimal stage should be? Yeah, so that's a great question because in networking, like with any other discipline that wants to use AI, there's not a lot of labeled data available. So what is the state I want to end up at? What is a problem state or what is a good state? All of this is labels that someone has to enter and that's not available at scale. And we are never going to be able to get it at the scale we want it. So one of our secret sources, if you will, is semi-supervised learning. But basic idea is that we are taking a lot of domain knowledge and using that domain knowledge to figure out what should be the right features for these models so that we can actually train these models in a scalable fashion. If you just throw it a lot of data at an AI model, it just does not converge. So how do I construct the features? And the other thing is how do I actually define what are good kind of end state conditions? What's a good network? And that's coming from domain knowledge too. That's how we are making AI scale for this domain. I mean, overall, I would say as you look at that, some of the parameters in terms of what you want to achieve are actually quite obvious, right? I mean, things like fewer drop calls for a cellular network. I mean, you know, that's good. And so figuring out what the metrics need to be and what the tuning needs to be for the network, that's where Ohana comes in in terms of their IP. All right, so, Shaker, give us a little bit of an understanding as to where this fits into the networking portfolio. You know, we heard from Patty a year or two ago, you know, what a strong, you know, push. Networking is, and the NSX number speaks for itself as to what's happening with that portfolio. No, absolutely. In fact, what we're doing here is actually broader than networking. It's sort of very pertinent to the network of a carrier, but that is a bulk of their business, if you will. I think if you sort of go back and look at VMware's, any, any, any vision, this is the notion of having any cloud and any application land on any cloud and then any device connected to those applications. On that any cloud side, we are looking at particularly two cloud pools, one which we call the telco cloud and the other is the edge cloud. And both of these fortuitously are now becoming sort of transformed in the context of 5G. So in one case in the telco cloud, you're looking at their core and access networks, their radio networks, all of this getting more cloudified, which essentially leads to a greater agility in service deployment. And then the edge is a much more distributed architecture, many points over which you can have compute storage, network management and security deployed. So if you now think about these sort of thousands of nodes and virtualized clouds, it is just impossible to manage this manually. So what you do need is greater, I mean orders of magnitude greater automation in the ability to go and manage an infrastructure like this. So with our technology now enhanced by uhana in that network portfolio in the telco edge cloud portfolio, we're able to go back to the carriers and tell them, look, we're not just foundational infrastructure providers, we can also then help you automate, help you get visibility into your networks and just help you overall manage your networks better for better customer experience and better performance. So what are some of the use cases that you see is being enabled by 5G? There's a lot of hype about 5G at the moment and not just 5G or so, things like Wi-Fi 6. It would appear to me that this kind of technique would work equally well for 5G or Wi-Fi or Wi-Fi 6. So what are some of the use cases you see these service providers or telco edge clouds using this for? Yeah, so I think overall, first of all, I'd say enterprise use cases are going to become a pretty prominent part of 5G, even though a lot of the buzz and hype ends up being about consumers and how much bandwidth and data they can get or whether 5G can pass through trees or not, right? But in fact, things like on-premise radio and whether that is private LTE and it's 4G or it's 5G, these are the kinds of use cases that we're actually quite excited about because these could be deployed literally today. I mean, sometimes they're not regulated. You can go in with existing architectures. You don't need to wait for standardization to break open a radio architecture. You could actually do it. And so this sort of going in and providing connectivity on an enterprise network that is an enhanced state of where it is today. We've already started that journey, for example, with Wello Cloud and branch networking. Now if we can take that to a radio-based architecture for enterprise networking. So we think a use case like that would be very prominent. And then based on edge architectures, distributed networks now becoming the next generation CDNs as an example. That's another application that we think would be very prominent. And then I think for consumers just sort of getting things like gaming applications off of an edge network. Those are all the kinds of applications that would consume this sort of high-scale reliability and performance. Sachin, can you give us a little sketch of the company pre-acquisition? You know, is the product all GA? How many customers? Or can you say what you had there? Sure, it was roughly three years old, the company itself. So relatively young. We were around 33 people total. We had a product that was already deployed with Tier 1 telcos. So it is in production deployment with the Tier 1 telco and is in production trials with a couple of other Tier 1 telcos. So we built the platform to scale to the largest networks in the world. And if I were to summarize it, we basically can observe, make sense in real time about every user in the network, what their experience is like and actually apply AI models on top of that to optimize each user's experience. Because one of the vision we had was the network today is optimized for the average. Whereas all of our web experience is personalized. The network experience is not personalized. Can we build a network where your experience is personalized for you, for the applications you are running on it? And this was kind of a foundation for that. I mean we, in fact, as we've been deploying our telco, cloud and carrier networks, we've also been counting roughly how many subscribers are being served up. And today we have over 800 million subscribers. And in fact, I was talking to someone and we were talking about that as being over 10% of the population of the world is now running on a lack of VMware infrastructure. And then along comes Johanna and they can actually fine tune this data right down to a single subscriber. So now you can see the sort of two ends of the scale problem and how we can do this using AI. So it's pretty powerful. Excellent. So if we have any problems with our service fighters, I'm sure we'll be able to give you a call. I'll be the tech support and I'll send Sachin over. I'd love to hear from both of you, what this acquisition means for the future of the PlayStation? Obviously VMware, global footprint, a lot of customers and resources, but what's it mean to your team and your product? I mean, definitely accelerating how quickly we can now start deploying this in the rest of the world. We as a small company have very focused on a few key customers to prove the technology. We have done that. And I think now it's the phase to scale it and repeat it across a lot of other customers. But I think it also gives us a broader canvas to play with. So we were focused on one aspect of the problem which is around, if you will, intelligence and subscriber experience. But I think with the cloud and with the orchestration products that are coming out of VMware, we can now start to imagine a full stack that you could build a network, a full carrier network out of using VMware technology. So I think it's a more exciting opportunity for us to be able to integrate not just the network data but also other parts of the stack itself and build a bigger product. It strikes me that this probably isn't just limited to telcos either, that service providers and carriers are one aspect of this, but particularly with 5G and things like deployments into factory automation systems. I can see a lot of enterprises starting to become much, in some ways, a little bit like a telco and they would definitely benefit from this kind of thing. Absolutely. Yeah, I mean, in fact, that's the basis of us internally even bringing our telco and Edge and IoT together in a common infrastructure pool. And so we're looking at that as the capability of deploying this type of technology across that. So you're exactly right. All right, Shaker, I want to give you the last word. Telco space, obviously the broader cloud has been a large growth area. What do you want people taking away from VMworld 2019 when it comes to your team? Yeah, I think to me, telcos have a tremendous opportunity to not just be the plumbing and networking providers that they can in fact be both the clouds of tomorrow as well as the application providers of tomorrow. And I think we have the technology and both organically as well as through acquisitions like Ohana to take them there. So I'm just super excited about this journey because I think while most of the people are talking about 5G as this wave that is just beginning for us, it's just a perfect coming together of many of these architectures that is going to take telcos into a new world. And we're super excited about taking them there. Well, Shaker, thank you so much for joining again. Sachin, congratulations. And good luck on the next phase of you and your team's journey along the way. Thank you. Thank you. For Justin Warren, I'm Stu Miniman. Stay with us, still a bit more to go for VMworld 2019. And as always, thank you for watching theCUBE.