 From Las Vegas, it's theCUBE. Covering VMworld 2018. Brought to you by VMware and its ecosystem partners. Hello everyone, we are here live in Las Vegas for theCUBE's exclusive coverage for three days. VMworld 2018, I'm John Furrier with my cohost, Stu Miniman. Our next guest is Andy Paxolstein, who's the founder and chief development officer and chairman of Arista Networks. More importantly, he's also the co-founder of Sun Microsystems. Invest in Larry and Sergey when they're in their PhD programs, legend in the industry, great to have you on, super excited to have you join this conversation. A pleasure to be here today. So first question is, besides all the luminary things you've done in your career, what's it like working with Jay Shree at Arista? Well, I actually met Jay Shree 30 years ago when she was at AMD selling us FTDI chips at Sun Microsystems. So I guess this dates both of us, but I've worked with her of all the years when I was at Cisco, obviously, and then we both started at Arista in 2008. So we've both been there now for 10 years together. In fact, the 10 year anniversary is coming up next month. Jay Shree's a great CUBE alumni. She's an amazing person, great technologist. We missed her, we should be here having more conversations with us on the CUBE. But stepping back, over your career you've seen many ways of innovation. You're involved in all of them, big ones happening, semiconductor, computers, and now with Arista going forward and now cloud, did you know the rocket ship of Arista was going to be this big? I mean, you designed it out at the beginning. What was the issue you were scratching in? Did you know it was going to be a rocket ship? Well, we had some very early feet. What led to the founding of Arista was we had lunch with our best friends at Google and Larry himself told me that the biggest from the head was not servers, but actually the networking and scaling that to the future size of the data centers. And they were going to go off to build their own network products because there was no commercial product on the market that would meet that need. So we thought with the emergence of merchant silicon we could make a contribution there and the focus of the company was actually on the cloud networking from the very beginning even though that wasn't even well in the student industry as being a major opportunity. So when we shipped our first products in 2009, 2010 many of them, we had some business on Wall Street Law latency but the majority of the opportunity was always the cloud. It's interesting you mentioned the Google and Larry and Sergey met Larry in particular about that time in history. You go back and look at what Google is doing at that particular time and now what they talk about at Google Cloud they were building their own large scale system and a massive scale involved. Yeah, they had about 100,000 servers in the early 2004 before they were in public. Now they have, who knows how many millions, right? And all of course the latest technology now. So the sheer size of the cloud, the momentum the cloud has, I think was hard to forecast. We did think there was going to be a shift but the shift was in fact more rapid than we expected. Yeah, Andy, you talked about cloud networking but today we still see there's such a huge discrepancy between what networking is happening in the data center and the networking that's happening in the hyperscalers. At this show we're starting to hear about some of the multi-cloud, you had some integrations between Arista and VMware that are starting to pull some of those together but maybe could give us a little bit about what you're seeing between the data center and the enterprise versus the hyperscalers when it comes to networking. So the data enterprise has still largely what we would call a legacy approach to networking which dates back 10, 20, 30 years and many of those networks are still in place and progressing very slowly but there also are enterprise customers who want to take advantage of what the cloud has done in terms of cloud networking including the much further scalability, the much further resiliency, the much greater automation. So all these benefits do apply equally well to the enterprise but it is a transition for customers to fully embrace that. So the work we're doing together with VMware on integrating our cloud vision, our physical switches with the micro segmentation is one element of that. But the bigger topic is simply an enterprise that wants to move into the future really should look at how did the cloud people build the networks, how can they run a very large data center with 10 network admins instead of hundreds of people and especially the automation that we have been able to provide to our customers automatic updating of software, being able to bring out new releases into a running network without bringing the network down. Nobody could even think about doing that 10 years ago. Yeah, you bring up a great point about automation. I love it in the keynote this morning, Pat Gelsinger talked about what was it? 39 years ago he did something at Intel, said we're going to do AI. Didn't quite call it AI back then, but he said, and now we're starting to see the fruits of what come out. In the networking world we've been talking about for decades automating the network more. You live through the one gig, 10 gig, 40 gig, 400 gig you're talking about. Are we ready for automation now is now that moment networking? I think people are ready for 30 years but the one thing is there always was a control plan in the network, the routing protocols, but for management there was never really a true management plan, meaning the legacy way is you dial in with SNMP into each switch and configure your access list manually more or less, and that's really a bad way of doing it because humans do make mistakes. You end up with inconsistencies and a lot of network outages actually has been traced to literally human mistake. So our approach with what we call cloud vision which is a central point that can manage your entire base of our risk to switches in the data center, it's all automated. You want to update a thing, you push a button and it happens and there's no more dialing into SNMP into individual switches. How would you advise people who are looking at the architecture of the cloud, who are replatforming large enterprises have been legacy all day long. You mentioned earlier just now on theCUBE that how the cloud guys were laying out the network was fundamental, how they grew. How should and how do people lay out the networks for cloud today? How would you, how do you see that? So the three big things that happened was merchant silicon has taken over because it's quite frankly much more scalable than traditional chips. And that's just the hardware, right? Then the leaf spine architecture that really our customers pioneered but is the standard in the cloud. It is, you know, use ECMP for load balancing, it works. It's the most resilient, the one thing, the single most important thing in the cloud is no outages, no downtime. You know, the network works no excuses, right? And our customers tell us that, you know, with our products and the leaf spine approach, you know, they have a better experience in terms of resiliency than any other vendor. So that's a very strong endorsement and that's as relevant to an enterprise customer as to a cloud customer. And then the automation benefit. Now to get the automation benefit, you know, you have to standardize on a new way of doing it. That's true. But it's just such a reduction in complexity and simplification. You know, you can actually look at this as an opaque saving opportunity, quite frankly. And, you know, in the cloud, they wouldn't have it any other way. They couldn't afford it, right? They have very large data centers and they only could operate these things in a fully automatic fashion. Andy, I want to get your reaction to what Pat Gelsinger said on stage this morning. He said, in the old days, the PEMFAR are phrasing, the network would dictate what the applications could do. It would enable that. We saw an enabling capability. Now with cloud, the apps can program the network. I'm paraphrasing that. As networks become more programmable and no outages, he made a quote. He said, the old adage was the network is the computer. The new adage is the application is a network. Okay, so what's your reaction to those things? Let me try to translate this. Sounds like an old sun slogan, doesn't it? That translates that for us. So the virtual networking, the NSX environment, which provides security at the application level, right? It's the natural way to do network security, because you really want to be as close to the application as you can physically be, or virtually be, which is right in the VM environment. So VMware clearly has the best position in the industry to provide that level of security, which is all software, software-defined networking. You do your security policies at level. Where we come in is with Cloud Vision now, we have announced a way to integrate with NSX micro-segmentation such that we can learn the policies and map them back down to the access list of the physical network to further enhance that security. So we don't actually create a separate silo for yet another policy management. We truly operate within their policy framework, which means you have the natural segmentation between the security engineers, which manage the security policies, and networking engineers that manage the physical network. They're highly optimized for the environment. Which actually works. So is that what you call macro-segmentation then? Well, we used to call it macro, but we actually know part of their micro-same because we truly learn their policies. So if you update a policy, it gets reflected back down to Cloud Vision and your physical networks, and it applies to physical switches, physical assets, physical servers, mainframe storage, whatnot, right? So it's a very smooth integration, and we think it's a demo at this point, but it will work and it's an open framework that allows us to work with VMware. Let me ask you a personal question. Looking at the industry, even look back at history as an illustration, TCPIP opened up, remember the old OSI stack that everyone tried to do that. TCPIP opened up so much on networking, internet working. Is there a technology enabler in Cloud that you see that's going to have that kind of impact? Is it NSX? How are customers going to deal with the multiple clouds? I mean, is there an interoperability framework coming? Do you see a real disruptive technology enabler that'll have that kind of impact that TCP spawn massive opportunity in wealth creation and startups and functionality? Is there a moment coming? So TCP, of course, was the proper layering of a network between the physical layer, layer one, layer two, and the routing or the internet layer, which is layer three. And without that, you know, this is a back to the old end to end argument. You know, we wouldn't have what we have today on the internet. That was the only rational way to build an architecture that could actually, and I'm not sure if people had a notion, you know, in 1979 when TCP was invented that it would become that big. They probably would have picked a bit bigger address space. But it was not just the longevity, but the impact it had was just phenomenal, right? Now, and that applied in terms of connectivity and how many things you have to understand about network to talk from point A to B. The NSX level of network management is a little different because it's much higher level. It's really a management plane, back to the point I made earlier about management planes, that allows you to integrate a cloud on your premise with one at Amazon or at IBM or in the future, Google and so on, in a way that you can have full visibility and you see exactly what's going on, all the security policies like. This has been a dream for people to deliver, but it requires to actually have a reasonable amount of code in each of these places, both on your server. It's not just a protocol, it's an implementation of a capability, right? And VMware NSX is the best solution that's available there and that I can see for that use case, which is going to be very important to a large number of enterprises, many of which want to have a smooth connection between on-premise and off-premise, and in the future to Edge, Telco and other things that don't even run a VM environment today, but that will allow them to be fully, securely linked into such a network. So you see that as a leading product, connect. It's definitely a leading product. When they have the most customers, the most momentum, the most market share, there isn't anything even close in terms of the, call it the software defined networking layer, which is what NSX implements, and we are very proud to partner with them at the physical layer to interact with their policy. Do you think that's going to have an impact of accelerating the multi-cloud world? Yes, because the whole point about multi-cloud is it has to be sort of vendor independent or I don't know, vendor neutral. You are going to see solutions from Amazon and Azure to bring their own sort of public cloud into the premise, but that only works with their backend, right? So there will be other offerings there, but in terms of true multi-cloud, I don't see any competition. Andy, I'd love to get your viewpoint on the future of Ethernet. I hear so many people the last few years, it's like, well, on the processor side, more laws played out. We can't get smaller. On the Ethernet side, there's not going to be the investment to be able to help get us to the next generation. There's limits in the technology. You've lived through so many of these architectural changes. Are we at the end of innovation for Ethernet? No, not at all. So my history with Ethernet dates back 40 years. So I worked on the first three megabit Ethernet in Syrox Park still, and then it was 10 megabit, 100 megabit, gigabit, 10, 40, 100, and now 400 coming out. So, see, Ethernet speed transitions are really just substitutions of the previous layer to technology, meaning, assuming they're more cost effective, they do get adopted very quickly. Of course, you need the right optics, you need the right equipment, but it's a very predictable roadmap, meaning that it's not like adopting a new protocol, but it's just faster. And more bandwidth cost efficient. So we are on the verge of 400 gigabits becoming available in the market. It will really roll out in any kind of volume next calendar year, and then even bigger volume in 2000. But in the meanwhile, 100 megabit Ethernet, excuse me, 100 gigabit Ethernet is still the fastest growing thing that industry has ever seen. It went from a million ports back in 2016 to, call it, five million ports last calendar year, expected over 10 million ports this year, expected 20 million ports next year. I mean, this is a speed of adoption that's unheard of, and we are at a risk that we are fortunate enough to be actually the market leader in 100 gigabit adoption. We have shipped more 100 gig ports than any other vendor, including Cisco for the last two years. So our ability to embrace new speeds and bring new technologies to market is, I would say, unparalleled. We have a very good track record there, and we're working really hard, you know, sort of burning the midnight oil to extend this to the 400 gig era, which is going to be another important upgrade, especially in the cloud. I should mention that the cloud is the early adopter of all the higher speeds. This has been during the 100 gig, will be more than 400 gig. I'm not sure too many enterprises need 400 gig, but the cloud is ready to get going as soon as it's cost-effective. Andy, for the folks that are looking at this 20-year wave coming that we're seeing kind of cloud, and it's been talked about on stage, and here on theCUBE, oh, it's going to be a 20-year run, transforming infrastructure. What's in your mind's eye? What do you see as the most disruptive thing that people aren't talking about in networking? What's going to be some things that might happen in the next 10 years in your mind that might happen that people aren't really aware of, that might not see it coming? Any innovations on the horizon that you're excited about, or people might not expect? Yeah, well, the cloud trend is fairly predictable. I would say, you know, all the IDC, all the analysts have predicted that big numbers on adoption have been pretty spot on, and if you look at the annual growth rate for cloud adoption, it's 40, 45, 50, and more percent. Now, there's a good question, of course, how the big cloud vendors in the end will compete against each other. You've got Amazon as the biggest, Microsoft is actually growing apparently fast in Amazon right now, but they have some catching up to do, and Google working overtime to get bigger, and they may differentiate in terms of their specific focus. For example, Google has a lot of AI technology internally that they've used for their own business, and with TensorFlow and so on, that they're arguably ahead of others, and they may just bet the farm on AI, right, and big data analytics, and things like that, which are a very compelling business opportunity is for any enterprise customer. So the potential value that can be created deploying AI correctly is in the perhaps trillions of dollars in the next 10 years, but it probably doesn't make sense for companies, for most companies to build their own AI data center, because you need a huge capital expense, a huge, what hardware do you use? It's going to evolve very quickly. So that may be one of the classical cases where you want to actually start in the cloud, and the only reason we ever move it on site is to have a well-defined environment, right? So I would actually say it's the new applications that may start in the cloud, that haven't even rolled out it in volume, like AI, that may be the biggest change that people didn't expect. Final question, what's the future of Arista? Oh, we're just working really hard to be the best provider of products, making the best products for our customers, both for the cloud and for enterprise, and one thing I was going to mention about Arista is that people think they're selling network boxes, which is not as which we do, but the vast majority of our investment is actually software, not hardware. So with over 90% of our R&D headcount is in software, and so the right way to think about it is actually to be a software company, not really a hardware company, and the saying we have internally is that hardware is easy, software is hard, because it's actually true. Software is much, much harder than building hardware these days, and the EOS software is now well over 10 million signs of codes, you know, written by over thousands of man years of engineering, so it has been a tremendous journey we have been on, but we're still, you know, scratching the surface of what we can do. And the focus of the software, obviously it's a big sense, software defined is driving everything. What are the key focus areas on the software that you guys are looking at? What's the key priorities for Arista? Yeah, we have talked about extending our business, you know, beyond the data center, into the campus. We announced our very first acquisition recently, which is actually a Wi-Fi company, but I can guarantee you it's going to be a software defined Wi-Fi network, not a legacy controller based approach, right? For enterprise, right? And we're not that interested in the hardware, we're really interested in providing a better managed solution to our customers. A lot of IOT action on it. Andy, thanks for taking the time to come on theCUBE, really appreciate it. Great to meet you and have you on theCUBE. Great conversation here. This is theCUBE. I'm John Furrier, Stu Miniman, breaking down all the top coverage of VMworld 2018, getting the input and the commentary from industry legends and also key leaders in the innovation and cloud, networking, this is theCUBE. Stay with us for more after this short break.