 Live from Mountain View, California, it's The Cube at OpenStack Silicon Valley brought to you by headline sponsor, Mirantis. Here are your hosts, John Furrier and Jeff Frick. Okay, we're back everyone. We here live in Silicon Valley for The Cube. This is our flagship program. We go out to the events and expect to see the noise. I'm John Furrier, the founder of Silicon Angle. My co-host Jeff Frick, who heads up our new Cube Silicon Valley operation where we'll be going out to a ton of events, meetups, you name it. If it's good, we'll be there. If it's not, we won't be there. So that's a very simple program. My co-host Jeff Frick, our next guest, is Flo Liebert, founder and CEO of Mesosphere. Mesos is popular, you guys are rockin'. I got people tweeting me right now. I can't wait to hear the interview. So welcome to The Cube. Thank you so much. So, you know, we love to talk about Born in the Cloud. I know Lon had arranged to get you on the program. I'm glad to have you. Try to get you on a crowd chat we had with Evan Powell at Stackstorm. Just a bunch of geeks chatting, but really the issue is that the Born in the Cloud guys, the DevOps, is now going mainstream. Certainly all the stuff about OpenStack is IT related. Yeah, some service provider, but especially want to be Born in the Cloud with hybrid clouds. So all these issues come to bear because scale is a big issue right now. Scaling, reliability, having a developer environment that's robust, compatible with what developers want, which is infrastructure as code, in a complex environment. So I got to ask you what you're takin' is on all that and what you guys built and what your company's doing in that area, because certainly there's a lot of, quote, technical challenges under the hood. Of course. Yeah, so the origin of our technology was really at UC Berkeley, where it was co-authored by Benjamin Hintman. And what we did is we were running into massive scaling issues back at Twitter when I worked there first. And that was roughly beginning of 2010. And the site was like many sites among a lithic application. That means it was, in the case of Twitter, actually it was a Rails application. It was powered by MySQL database and had some memcached servers that were used as caches. And what Twitter needed to do at that time in order to scale up the infrastructure was turn a lot of the functionality into microservices. So for example, the first thing that you generally pull out, I think in sites like this, the search contains a lot of logic and usually it's pretty latency sensitive. So search was pulled out and a bunch of other components were pulled out into these microservices and we needed a way to easily deploy these microservices as they were being built. But also, the problem was when we had a microservice, it didn't necessarily take up the entire resources that were offered by one big server. So what we wanted to do is figure out a way to bin pack, to automatically bin pack a lot of these microservices onto these big servers. And we looked over to UC Berkeley where Benjamin Hindman was working on Mesos, brought him in for a talk and all the folks that were previously at Google that had joined Twitter saw some similarities in the technology that they were using at Google and it was called Borg back in the day. Later became Omega and they saw the similarity in it and saw that this can really solve a lot of these problems. The deployment problem, but also the orchestration problem of course and also increased utilization by two to three acts. So this is the big discussion everyone has that might not be inside the industry. All these large scale web scale companies are essentially building out while they're scaling. So we even know the early days of Twitter, we even know Zynga, they all start on Amazon. They bootstrap, they get capital efficient and boom, they're scaling and there's a kind of diminishing return where they got to set up their own gear. Stand up their own infrastructure and then there's a huge challenge because the applications are putting load in the system and the ab guys are buy more gear, buy more gear, provision this. So that's your experience where you've had this. Talk about the pain and how you guys looked at that and then where you are today. So yeah, one of the things we saw was that you could page every night, specifically the Airbnb we were built on top of Amazon and on Amazon, the platform is not necessarily as reliable as when you have your own hardware because Amazon might just turn off one of your servers overnight. So if you're a DevOps or an SRE, you get paged and then you have to deal with it and that was one of the big things that we actually wanted to fix and that's why we started to build Mesosphere. We wanted to automate a lot of these tasks that have to be done in your data center completely away for you. At the same time, we wanted to increase resource utilization. So one of the first projects we built at Mesosphere was called, or is called Marathon and it's a way to orchestrate containers at scale. So you can go in and you can say, I want to launch a Rails application that depends on maybe a memcached server and you can deploy this application topology and then just scale it up. And in fact, it's really easy to actually use auto scaling by hooking into our REST APIs and then just increasing the number of Rails instances, memcached servers and so forth. So let's unpack this nuance about horizontally scalable. That is the beautiful thing about the cloud and Amazon. It's horizontally scalable. As demand comes in, there's all kinds of tools. It's horizontally scale, load comes in, boom, it's done, it feels good, it's easy to do. Almost like pushing a button, I'm over simplifying. Versus standing up gear, which requires a lot of configuration management, policy-based everything, orchestration, integration, automation, which is a real pain in the ass. Straight up, right? So okay, so now we're in this horizontally scalable world. What are the table stakes for someone out there who's saying, hey, you know what? I want horizontally scalable. I want an app environment that at will I could program infrastructure. What do they do? Give us the playbook. Well, I think the first thing you generally do is you figure out how you package your application. And through the advent of the Docker file format, we now have a really good way of packaging an application up. And Google has actually put a lot of effort into isolation. So in 2007, 2008, isolation came into the Linux kernel and Google contributed heavily. And that isolation can now be used in order to run these applications next to each other on the same box. And that's, by the way, one of the key features that we take advantage of in Mesos and Marathon when we want to scale up your application. We can scale it on the same box, but we can also scale it across numerous servers. So talk about the scaling containers at scale. It sounds kind of mouthful, but what does that mean? Docker's hot right now because it really is a nice framework. Certainly the developers adopted it. And they really cracked the code on how to use open source in a way that actually punishes you for not being open. But talk about that dynamic. Yeah, so I think Docker is great. I mean, they've really invented a great file format. But I think there's much more to it. If you want to run a large data center at scale, you need to orchestrate. And orchestration is something that's actually really difficult to do. It took Google years of hardcore development to get to a point where they could scale up 2,000s, 10,000s, 100,000s of servers. And Twitter likewise, it took them a lot of time and engineering effort to get to a point where Mesos was as stable as it is today. Where you can actually go in and provision and then run on a 10,000 node cluster. So I got to ask you the question. I'll see you looking at your investor list, Andreessen Horowitz, the list goes on and on. Very disruptive investors. They like to go after the disruptors. You are being a disruptor. Who are you disrupting? And who's disrupted by you guys? So I think that's a great question to ask. So I think to a certain extent right now, the whole virtualization space is being disrupted. And I think it's being disrupted in a way that in the past, we always thought about bin packing onto a single server, multiple applications. And back in the day, 10 years, 15 years back in the height of the client server era, the applications were relatively small and the servers kept increasing in capacity. At most, you had maybe an Oracle server with a slave, with a slave instance, and it spanned one server and the slave server was also one big server. But oftentimes, you had a couple of Apache web servers and so forth on the same box. And there we used virtual machines in order to manually bin pack the server to increase utilization. And what we are seeing now is applications today are from day one distributed systems. They no longer live on just one box. They are examples of that are Hadoop, Spark, Storm, and Cassandra, and MongoDB. And many more. They run on multiple servers from day one. And in that world, we actually need a different model. We think we need an aggregation model, not a virtualization model. And you could think of it as we're virtualizing the entire data center to make it look like one big computer. And we're in the business. That's the mega trend right now. It's not going to be siloed into a box. And we want to build the operating system that runs on your entire data center. Yeah, so where are you with that? So where are you in the stem? That's a lofty dream, great goal to have, make it look like one monolithic resource, but there's a lot of stuff moving around with the distributed computing architecture. So large scale computer science is at play here. What's under the hood? I mean, what is the key secret sauce? So really the key secret sauce I think is Mesos for us. So Apache Mesos is a project that has really figured out how to represent resources that each of the nodes, that each of the nodes provides and hand them up to the applications. And then as an application developer, you can actually go in and say, I'm programming against resources rather than against individual machines. And I think that's really powerful. And those resources can shift and go where there's more opportunities for capacity, et cetera. Exactly. And you can on a global level decide a policy, which application should get how many resources. So that makes for really interesting dynamics in your data center. This is the trend we were talking about at VMworld, Jeff, the dictating the policy to the infrastructures, the applications versus the other way around. This was not the way it was when I was growing up in the business. You stack a bunch of gear and they were limited by the engines that you had. That's a completely different shift. Yep. So I got to ask you a question for the folks out there that are learning about you guys. Certainly the buzz is good. You guys are a great track record team. What are you and what aren't you? So if you had to put a statement out there, this is what we are. We're more like, we're the Uber of this, not the same owner. I'm just kidding. That's just an inside joke, Silicon Valley. So, but what are you and what aren't you? We're the operating system for the data center. And what are we not? Well, everything else. We're not a big data company. Let's just say you're the Uber for the data center. Just go with that. So it'll definitely get your next round done. You've got to be the Uber of something these days. No, we're not really in the big data space. We're really in the infrastructure space. And so we're an infrastructure company, yeah. So Flo, you've got a lot of history with kind of born in the cloud applications and really big single applications that's getting really big with Twitter and Airbnb. As you talk to enterprises, talk about the challenges that they have where it's really many, you know, trying to apply these principles across many applications in the same data center, as well as when they've got applications that weren't necessarily built this way from the ground up to really be able to assign resources as a service versus how they were pretty much rigidly signed. Are people trying to convert those? You know, can they leverage some of these lessons learned in that environment? So that's really the beautiful thing about a marathon which we've written. It sort of acts like a shim between your existing applications which you don't have to modify, that run on Linux today, that you can run on Amazon today, and the Mesos system. So we've written a marathon specifically for that purpose. You can take any existing Linux application today, deploy it at scale, and actually get self-healing and higher resource utilization as a nice side effect. Wow, so you can apply it to your legacy stuff. Oh, absolutely, yes. So VMware hopes to be the operating system of the data center, software defined, data center is the hot buzzword. I mean, that's in essence what you're saying. Is that correct? I'm sure there can always be a lot of players in a big space. Okay, so you're disrupting who again? VMware, IBM, HP, are they friends, foe, frenemy? I think we're really filling a void. They are, you are. We are. They're voided, so yeah. But this is the big thing here, OpenStack is really about IT. So how do you come into the IT world? Have a little bit different requirements. They're born on premise, they're not born in the cloud, but they want to be born in the cloud in a half way. What, it's a basis, like a hybrid. So actually I think we play really nice with OpenStack. So a lot of our customers actually have started to invest heavily into OpenStack and it's really great for provisioning. So if you want the Amazon-like provisioning in your data center and you have already invested in OpenStack, you can just run Mesos on top of it. And then you get additional automation on top of your already highly automated data center. Now, that being said, there are companies that are starting with Greenfield projects and like Twitter for example, right? When it decided to move down this path of having everything run via Mesos, decided to not use any other technology and they're just using Linux C groups, so barebound containers alongside Mesos and some of their frameworks that they've written on top of Mesos directly. So talk about OpenStack. What do they need to do better? Obviously orchestration is a conversation that we have every time we go to an OpenStack show and it's kind of higher up the stack but there's still some critical issues. Automation again is another buzzword, but they seem to be stuck in the building block mode. What is happening from your perspective in the OpenStack community? I mean it's evolutionary, so it's not, they're not taking a step back, but some say they're not moving fast enough to provide that basic substrate infrastructure. Well I think OpenStack has set a really high goal also for themselves, so I think it's a tough problem and even just solving the provisioning piece I think takes a lot of time and I think they've done a great job on that, so I'm excited to see what comes next. Okay so we have a question from Stu Miniman who's watching from Wikibon, a big fan of yours, you can see him geeking out right now at his terminal. Can Flow explain how Mesos interacts with Docker and Kubernetes? Of course, so Kubernetes is a project built by Google and it was supposed to mimic some of the workflows, some of the developer workflows that Google has in order to roll out the applications and we went in and decided to partner with Google in order to bring Kubernetes on top of Mesos so that it can scale, so that it can scale massively because one of the pieces that Google has not open sourced that they are running internally is called Omega and was previously known as Bork and that is not available for everybody so that was a great opportunity for us to say hey, we're going to integrate the two projects and give Kubernetes fans the ability to run at massive scale. So are you a vitamin or an aspirin to Kubernetes? We're both, well we're both. I think you're a vaccine. No, I mean, we're really a vitamin. We really coexist nicely with them. And they're growing too. Yes, exactly. They're part of that thing. Now for Docker, to answer the second part of your question, so again, I think Docker has done a great job figuring out how to package an application with a Docker file format and if you just have a Docker container, you can't do really that much with it, right? So that's where we come in and we figure out a way how to orchestrate it at scale and not just for 10 notes, not just for 100 notes, but for thousands and tens of thousands of notes. Okay, so I want you to kind of put your messaging aside and put your visionary hat on and your geek hat on and paint the picture of what the data center looks like five to 10 years from now. CIOs, what's the preferred future look like for you? As the founder and CEO of Mesa Sphere, for the ideal future, what is it going to look like in the data center? I think there will be far fewer vendors who sell into the data center. I think it's going to consolidate quite a bit. I think- They'll have a hard time with that. They'll be clutching and grabbing until the last minute. Yeah, I think also more and more commodity components will make their way into the data center. So I think we're going to see a lot of chips that we're seeing in cell phones today, ARM and atom processors moving into the data center because they're just more power efficient and you can have many more processors in your data center that way. I think power virtualization is going to be a big aspect. I think right now, the racks that you're seeing, every node has its own power supply. I don't think that's actually, that's the future. I think we'll see a different rack format where we can just swap out, where we can just swap out disks that are by the way no longer spinning. They're all flash based and RAM and CPUs that are really small. So what about the role of data? Obviously the internet of things is big. Is this going to be just our native part of the infrastructure? Yeah, so I think I have to actually speak on behalf of Mesos here again. So I think specifically with the internet of things and the need for more data processing applications as that need arises, we will see more frameworks being built directly against the data center. And that's where Mesos really comes in and can help you because Spark, are you familiar with Spark? Yes, very much so, yeah. So Spark was one application that was built against the Mesos API initially and that's how they were able to move so quickly. Now, when they first started Spark, there was no Mesosphere and there were no Mesos RPMs and Debian packages. So when you wanted to run Spark, you also had to compile Mesos from scratch. And that's why they had to backfill a lot of the networking that Mesos gives you for free. A lot of the messaging, the RPC, the isolation, they had to backfill that. And what I think is like, we're not at the end of the tunnel of data frameworks, data processing frameworks. There will be many more. Flo, thanks for coming on theCUBE. The founder and CEO of Mesosphere, innovation is happening, it's just doing well. Keep an eye on these guys, certainly as the evolution of the stack continues. Congratulations, the data center is software-defined, it's happening. And we're here at the OpenStack Silicon Valley, should be right back after this short break.