>> Live from New York, It's the CUBE! Covering theCUBE, New York City 2018. Brought to you by SiliconANGLE Media and its Ecosystem partners. >> Welcome back to the live CUBE coverage here in New York City for CUBE NYC, #CubeNYC. This coverage of all things data, all things cloud, all things machine learning here in the big data realm. I'm John Furrier and Dave Vellante. We've got two great guests from Cisco. We got DD who is the Vice President of Data Center Marketing at Cisco, and Han Yang who is the Senior Product Manager at Cisco. Guys, welcome to the Cube. Thanks for coming on again. >> Good to see ya. >> Thanks for having us. >> So obviously one of the things that has come up this year at the Big Data Show, used to be called Hadoop World, Strata Data, now it's called, the latest name. And obviously CUBE NYC, we changed from Big Data NYC to CUBE NYC, because there's a lot more going on. I heard hallway conversations around blockchain, cryptocurrency, Kubernetes has been said on theCUBE already at least a dozen times here today, multicloud. So you're seeing the analytical world try to be, in a way, brought into the dynamics around IT infrastructure operations, both cloud and on premises. So interesting dynamics this year, almost a dev ops kind of culture to analytics. This is a new kind of sign from this community. Your thoughts? >> Absolutely, I think data and analytics is one of those things that's pervasive. Every industry, it doesn't matter. Even at Cisco, I know we're going to talk a little more about the new AI and ML workload, but for the last few years, we've been using AI and ML techniques to improve networking, to improve security, to improve collaboration. So it's everywhere. >> You mean internally, in your own IT? >> Internally, yeah. Not just in IT, in the way we're designing our network equipment. We're storing data that's flowing through the data center, flowing in and out of clouds, and using that data to make better predictions for better networking application performance, security, what have you. >> The first topic I want to talk to you guys about is around the data center. Obviously, you do data center marketing, that's where all the action is. The cloud, obviously, has been all the buzz, people going to the cloud, but Andy Jassy's announcement at VMworld really is a validation that we're seeing, for the first time, hybrid multicloud validated. Amazon announced RDS on VMware on-premises. >> That's right. This is the first time Amazon's ever done anything of this magnitude on-premises. So this is a signal from the customers voting with their wallet that on-premises is a dynamic. The data center is where the data is, that's where the main footprint of IT is. This is important. What's the impact of that dynamic, of data center, where the data is with the option of a cloud. How does that impact data, machine learning, and the things that you guys see as relevant? >> I'll start and Han, feel free to chime in here. So I think those boundaries between this is a data center, and this a cloud, and this is campus, and this is the edge, I think those boundaries are going away. Like you said, data center is where the data is. And it's the ability of our customers to be able to capture that data, process it, curate it, and use it for insight to take decision locally. A drone is a data center that flies, and boat is a data center that floats, right? >> And a cloud is a data center that no one sees. >> That's right. So those boundaries are going away. We at Cisco see this as a continuum. It's the edge cloud continuum. The edge is exploding, right? There's just more and more devices, and those devices are cranking out more data than ever before. Like I said, it's the ability of our customers to harness the data to make more meaningful decisions. So Cisco's take on this is the new architectural approach. It starts with the network, because the network is the one piece that connects everything- every device, every edge, every individual, every cloud. There's a lot of data within the network which we're using to make better decisions. >> I've been pretty close with Cisco over the years, since '95 timeframe. I've had hundreds of meetings, some technical, some kind of business. But I've heard that term edge the network many times over the years. This is not a new concept at Cisco. Edge of the network actually means something in Cisco parlance. The edge of the network >> Yeah. >> that the packets are moving around. So again, this is not a new idea at Cisco. It's just materialized itself in a new way. >> It's not, but what's happening is the edge is just now generating so much data, and if you can use that data, convert it into insight and make decisions, that's the exciting thing. And that's why this whole thing about machine learning and artificial intelligence, it's the data that's being generated by these cameras, these sensors. So that's what is really, really interesting. >> Go ahead, please. >> One of our own studies pointed out that by 2021, there will be 847 zettabytes of information out there, but only 1.3 zettabytes will actually ever make it back to the data center. That just means an opportunity for analytics at the edge to make sense of that information before it ever makes it home. >> What were those numbers again? >> I think it was like 847 zettabytes of information. >> And how much makes it back? >> About 1.3. >> Yeah, there you go. So- >> So a huge compression- >> That confirms your research, Dave. >> We've been saying for a while now that most of the data is going to stay at the edge. There's no reason to move it back. The economics don't support it, the latency doesn't make sense. >> The network cost alone is going to kill you. >> That's right. >> I think you really want to collect it, you want to clean it, and you want to correlate it before ever sending it back. Otherwise, sending that information, of useless information, that status is wonderful. Well that's not very valuable. And 99.9 percent, "things are going well." >> Temperature hasn't changed. (laughs) >> If it really goes wrong, that's when you want to alert or send more information. How did it go bad? Why did it go bad? Those are the more insightful things that you want to send back. >> This is not just for IoT. I mean, cat pictures moving between campuses cost money too, so why not just keep them local, right? But the basic concepts of networking. This is what I want to get in my point, too. You guys have some new announcements around UCS and some of the hardware and the gear and the software. What are some of the new announcements that you're announcing here in New York, and what does it mean for customers? Because they want to know not only speeds and feeds. It's a software-driven world. How does the software relate? How does the gear work? What's the management look like? Where's the control plane? Where's the management plane? Give us all the data. >> I think the biggest issues starts from this. Data scientists, their task is to export different data sources, find out the value. But at the same time, IT is somewhat lagging behind. Because as the data scientists go from data source A to data source B, it could be 3 petabytes of difference. IT is like, 3 petabytes? That's only from Monday through Wednesday? That's a huge infrastructure requirement change. So Cisco's way to help the customer is to make sure that we're able to come out with blueprints. Blueprints enabling the IT team to scale, so that the data scientists can work beyond their own laptop. As they work through the petabytes of data that's come in from all these different sources, they're able to collaborate well together and make sense of that information. Only by scaling with IT helping the data scientists to work the scale, that's the only way they can succeed. So that's why we announced a new server. It's called a C480ML. Happens to have 8 GPUs from Nvidia inside helping customers that want to do that deep learning kind of capabilities. >> What are some of the use cases on these as products? It's got some new data capabilities. What are some of the impacts? >> Some of the things that Han just mentioned. For me, I think the biggest differentiation in our solution is things that we put around the box. So the management layer, right? I mean, this is not going to be one server and one data center. It's going to be multiple of them. You're never going to have one data center. You're going to have multiple data centers. And we've got a really cool management tool called Intersight, and this is supported in Intersight, day one. And Intersight also uses machine learning techniques to look at data from multiple data centers. And that's really where the innovation is. Honestly, I think every vendor is bend sheet metal around the latest chipset, and we've done the same. But the real differentiation is how we manage it, how we use the data for more meaningful insight. I think that's where some of our magic is. >> Can you add some code to that, in terms of infrastructure for AI and ML, how is it different than traditional infrastructures? So is the management different? The sheet metal is not different, you're saying. But what are some of those nuances that we should understand. >> I think especially for deep learning, multiple scientists around the world have pointed that if you're able to use GPUs, they're able to run the deep learning frameworks faster by roughly two waters magnitude. So that's part of the reason why, from an infrastructure perspective, we want to bring in that GPUs. But for the IT teams, we didn't want them to just add yet another infrastructure silo just to support AI or ML. Therefore, we wanted to make sure it fits in with a UCS-managed unified architecture, enabling the IT team to scale but without adding more infrastructures and silos just for that new workload. But having that unified architecture, it helps the IT to be more efficient and, at the same time, is better support of the data scientists. >> The other thing I would add is, again, the things around the box. Look, this industry is still pretty nascent. There is lots of start-ups, there is lots of different solutions, and when we build a server like this, we don't just build a server and toss it over the fence to the customer and say "figure it out." No, we've done validated design guides. With Google, with some of the leading vendors in the space to make sure that everything works as we say it would. And so it's all of those integrations, those partnerships, all the way through our systems integrators, to really understand a customer's AI and ML environment and can fine tune it for the environment. >> So is that really where a lot of the innovation comes from? Doing that hard work to say, "yes, it's going to be a solution that's going to work in this environment. Here's what you have to do to ensure best practice," etc.? Is that right? >> So I think some of our blueprints or validated designs is basically enabling the IT team to scale. Scale their stores, scale their CPU, scale their GPU, and scale their network. But do it in a way so that we work with partners like Hortonworks or Cloudera. So that they're able to take advantage of the data lake. And adding in the GPU so they're able to do the deep learning with Tensorflow, with Pytorch, or whatever curated deep learning framework the data scientists need to be able to get value out of those multiple data sources. These are the kind of solutions that we're putting together, making sure our customers are able to get to that business outcome sooner and faster, not just a-- >> Right, so there's innovation at all altitudes. There's the hardware, there's the integrations, there's the management. So it's innovation. >> So not to go too much into the weeds, but I'm curious. As you introduce these alternate processing units, what is the relationship between traditional CPUs and these GPUs? Are you managing them differently, kind of communicating somehow, or are they sort of fenced off architecturally. I wonder if you could describe that. >> We actually want it to be integrated, because by having it separated and fenced off, well that's an IT infrastructure silo. You're not going to have the same security policy or the storage mechanisms. We want it to be unified so it's easier on IT teams to support the data scientists. So therefore, the latest software is able to manage both CPUs and GPUs, as well as having a new file system. Those are the solutions that we're putting forth, so that ARC-IT folks can scale, our data scientists can succeed. >> So IT's managing a logical block. >> That's right. And even for things like inventory management, or going back and adding patches in the event of some security event, it's so much better to have one integrated system rather than silos of management, which we see in the industry. >> So the hard news is basically UCS for AI and ML workloads? >> That's right. This is our first server custom built ground up to support these deep learning, machine learning workloads. We partnered with Nvidia, with Google. We announced earlier this week, and the phone is ringing constantly. >> I don't want to say godbot. I just said it. (laughs) This is basically the power tool for deep learning. >> Absolutely. >> That's how you guys see it. Well, great. Thanks for coming out. Appreciate it, good to see you guys at Cisco. Again, deep learning dedicated technology around the box, not just the box itself. Ecosystem, Nvidia, good call. Those guys really get the hot GPUs out there. Saw those guys last night, great success they're having. They're a key partner with you guys. >> Absolutely. >> Who else is partnering, real quick before we end the segment? >> We've been partnering with software sci, we partner with folks like Anaconda, with their Anaconda Enterprise, which data scientists love to use as their Python data science framework. We're working with Google, with their Kubeflow, which is open source project integrating Tensorflow on top of Kubernetes. And of course we've been working with folks like Caldera as well as Hortonworks to access the data lake from a big data perspective. >> Yeah, I know you guys didn't get a lot of credit. Google Cloud, we were certainly amplifying it. You guys were co-developing the Google Cloud servers with Google. I know they were announcing it, and you guys had Chuck on stage there with Diane Greene, so it was pretty positive. Good integration with Google can make a >> Absolutely. >> Thanks for coming on theCUBE, thanks, we appreciate the commentary. Cisco here on theCUBE. We're in New York City for theCUBE NYC. This is where the world of data is converging in with IT infrastructure, developers, operators, all running analytics for future business. We'll be back with more coverage, after this short break. (upbeat digital music)