 This is a return trip for me to China. We came here in May after we launched DCOS as an open source project. And I'm excited to see how much more enthusiasm and innovation there is this time, now that we're back for Mesoscon Asia. So thank you to everybody for joining us these two days. Thank you to the Linux Foundation for helping us put this event on. And thanks to all of our partners, including IBM, who are here supporting Mesos and Mesoscon today here in China. My talk has a bit of a strange title. It probably makes more sense in English than it does if you try to translate it. But what I'm going to be talking about is basically how Mesos, as a core technology, is fantastic. But it's never just Mesos that makes a data center great that makes modern applications possible. There's always more to it than that. And so I'm going to talk about what some of those other projects are that typically get run with Mesos. And I'm going to talk about why we created DCOS as an open source umbrella that brings together a bunch of these projects into a single distribution. So the whole point of this talk is to help give you a little bit of context outside of Mesos. I think sometimes it's easy at Mesoscon to just focus on Mesos. And we want to make sure that everybody is talking about all the other technologies that work well with Mesos, that were designed to run with Mesos, and how those things kind of fit together. So because we're running a little late, I don't know that I'm going to have time at the end for questions. But I'll be here all day. And I'll be at the Mesosphere booth outside. So if you have any questions or if you want to talk further, don't be shy. You can reach out to me. All right, let's get started. So who am I? I am the format of the slide is a little funny. So apologies for that. That's supposed to be a circle. My head's not quite that wide. I'm Aaron Williams. I head up our global community at Mesosphere. And that's two things. One, I'm responsible for DCOS as an open source project. And we'll talk a little bit more about what that means in a minute. And second, I run a team of advocates who go around the world and tell people about Mesos and about DCOS and help them take advantage of and adopt these technologies. So primarily, my team is responsible for one thing. And that's driving adoption of DCOS and Mesos around the world. My background, I have a background in communities. So I ran the Java community process for Sun Microsystems. I ran communities at SAP. So I've worked at large companies running their very large developer communities. And I've also spent time at startups. So I was CEO of two startups and really enjoy sort of both sides of that spectrum. I like being at really small companies and I like being at really big companies. All right, so let's get started. I almost always start my presentations with this quote. And it's not just because Satya's a super smart guy and I love all the things they're doing at Microsoft these days. It's impressive to me how they've really turned that company around. I think just the announcements that came out this week about them joining the Linux Foundation is a big step in the right direction. But I start with this quote for a similar reason. It helps us remember that there's some business value to the decisions that we're making with Mesos and with DCOS. This isn't just about making the right technical decision. It's also about the reason why. It's the business reason why we make these decisions. You know if you're a big company like Microsoft you can afford to have a team of engineers that are customizing a bunch of components in your data center to make it perfect for your environment and for your applications. Most companies though don't have that luxury. They have a much smaller team of engineers that are working on building out their applications and their data center. And so what this quote reminds us is that you should always be focused on that business value because your data center now is a competitive weapon against your competition. You're not just competing on the quality of your products. You're competing on the quality of the software that helps you make those products. When Ben Heimann showed the slide earlier about the different industries that are adopting Mesos the reason that is such a diverse set of industries is because those companies see value in being able to have Mesos as their underlying technology but they are competing on the value of the software they're building on top of Mesos. So as I go through my presentation I'm gonna refer back to this quote because it'll help to remind us of why we make some of the decisions that we make when we build out our data center. All right, so a little history here. And I suppose everybody in this room knows this history so I'll go very quickly through it. But it's really a story of increased complexity going from a single mainframe to multiple servers to virtual machines to now containers inside virtual machines. You increase the sophistication of your data center. You increase the complexity of your data center. And now we're at a point where it gets too difficult to manage each of these individual pieces even. So how do you start to think about your data center differently? It's not a collection of individual components anymore. Now it's one big pool of resources that you can run your applications on top of. And that's really where Mesos comes in. So this is the story of sort of how we got to this new paradigm of having a data center form factor. It's not about caring about what's happening on each of the individual machines. It's about caring what's happening at the data center level. Our CEO sometimes uses the analogy of when you run a web browser on your laptop you don't care what core it runs on. Of course you don't. You don't actually pay attention at that level. You care about it at the laptop level. And I think this is very similar now where we're at today. We care about our data center at the data center form factor and we care about running applications across that entire pool of resources. So what that means is we used to have sort of a traditional approach where things were pretty siloed. You ended up with these sort of individual services running on individual data centers or individual stacks of servers. And that worked but it really didn't give us the ability to have that data center form factor. We weren't thinking about things at that level. We were still thinking about things at the individual machine level. Now when we think about the moderate approach what we have is the ability to pool all of these resources with a single distributed operating system. That enables us to run any kind of workload on top of that data center at that form factor. So this is where we see the future. Where we see modern applications, the companies that we're working with to adopt Mesos and adopt DCOS. This is the architecture that they're adopting. And I think we even saw it this morning in some of the keynotes. If you see their architecture diagrams they look a lot like that, not like this. And this is why Mesos has become a champion. This is why Mesos has become the technology that it is widely adopted, used at very large scale, trusted by very large companies who are betting their business on using Mesos as their core technology. We know all these name brands, right? We know all these companies, we know what they're doing. You're gonna hear from some of them today and tomorrow and about their experience using Mesos. But the interesting thing for me, when we went to talk to these companies and learn about how they were using Mesos, the way they were using it, what their architecture was, it became really clear that none of these companies are using just Mesos. They don't just deploy Mesos and then they're done. They have to figure out how are they gonna do container orchestration? How are they going to do service discovery? How are they gonna do storage and networking? All of these pieces now, once you've got your data center form factor, once you're thinking about your architecture at that level, there's still a lot of problems that need to get solved. And so I think that's really where this talk comes in because all of these companies had to look around and look for technologies that work well with Mesos and had to make hard decisions about what technologies were they going to use with their Mesos cluster. And they all ended up with a very similar architecture. And I think at a high level at least, this gives us sort of the starting point of thinking about what these other technologies are. So I'm just gonna talk for a second about what that architecture looks like. We call this the architecture of the modern application. So I showed you what a modern data center looks like. Now let's talk about it at the application level and thinking about it from the view of the application. So what we believe modern applications are containers with microservices, but paired with just as important as the big data analytics engines, CICP, other things that you want to run in your data center. It's not just about having containers. None of those companies I showed before are just running containers. Containers are important. They're critical. But it's also about the other things that you need. We heard this morning talk about Spark, talk about different container orchestrators. And so what we see is an opportunity then to take this architecture, build out all the complete pieces that are needed to be able to make that architecture possible. Those things I mentioned before, service discovery, gluing and CLI, the interface to the data center. And if we do this right, then you could run this on any infrastructure. So you could run it in the cloud. You could run it on-prem. You could run it in a hybrid environment. And you wouldn't have to change anything about your modern application. That would run anywhere. Now, let's think back to what I was saying before about Satya's quote. If we think about building an architecture like this, which of these pieces are the pieces that will have the biggest impact on my modern application's ability to compete against my competitors? The biggest impact is right here. Because this is the stuff that's unique to your application. Your microservices, your containers, the things that make your application unique, your business logic. The reality is, these big data, these analytics engines, they're pretty common. There's a lot of them, and that's good. We want choice, but very few people are actually building their own analytics engine or their own big data pipeline. Same thing down here. When we think about the components that are needed, service discovery, load balancing, the GUI, the CLI, these are pieces where, of course, you could go build your own if you wanted to, but why would you invest your time and energy there when it's not going to have a big impact on the competitiveness of your application? So what we encourage people to do is invest as much time and energy as possible in this part of the architecture and use as many existing pieces as you can down here and over here. Take advantage of the fact that we have really vibrant open-source communities that are building out those components. Take advantage of the fact that you can contribute and participate and engage in those communities, but you don't have to invest your resources, your very scarce resources in building that out yourself. Now, just to be clear, if you're a company like Apple, there are things in here that are absolutely competitive to you. There are things that differentiate Apple and they go and build their own, and that's okay. We want to encourage that to happen, that's a good thing. But most companies aren't Apple, right? Most companies don't have a gigantic, donut-shaped campus in Silicon Valley full of engineers. So, what we say is make sure you're using your resources to get the most competitive advantage out of your architecture. And for most companies, that's gonna be investing in the microservices and containers. And if we do this right, if we all get together and we focus on this kind of an architecture, then we will have that healthy ecosystem of frameworks, that healthy ecosystem of partners that are working with us to make this picture complete. That's really, again, what the whole point of this talk is about, is being able to think about your architecture where mesos is an important piece, but it's just a piece. So, this is where DCOS comes in, and I have to admit, I'm very biased, right? I run the DCOS project, okay, fair enough. But, what I'm gonna do is show you why I think, or sort of how we came to the decisions that we came to within the DCOS project, and maybe that fits with some of the challenges that you're having as well. If not, of course, you could recompose these pieces in a way that's more appropriate for you. But I think what you'll find is that the DCOS project does a good job of bringing together the core components that are needed, makes it easy for you to install, easy to get started. And because it's an open source project, if there's something in there that we've forgotten or something that doesn't fit for you, you're always welcome to come and join the community and help us make it better. Okay, so what does the modern application then look like on top of DCOS? Well, going back to that previous slide, these are all the components that were in that purple box, right? We've got a GUI and a CLI. We give you the ability to manage the entire data center at the data center level, either through a GUI or through a CLI, depending on your preference. We have Mesos as the core underlying technology, of course, no surprise there. We have a service manager called Marathon, which is great for running containers and all kinds of containers. We have also the frameworks. You can install your favorite frameworks, analytics, big data, fast data, et cetera. And then we have what's called the universe, which gives you an easy one-click or one-command way to install these frameworks into your data center, to make these distributed frameworks available across your entire data center. How many people here have tried to install Spark across a data center? Very few. Okay, so you have to trust me. It's a pain. It's very painful. And that's not because there's anything wrong with Spark. Spark is great. I love Spark, but making it work inside a distributed architecture can be challenging. But if you've defined these pieces underneath it, it becomes a lot easier. It becomes possible to run Spark with that single command or that single click to make the Spark framework run in your distributed system. And it's not just Spark. It's Kafka. It's your favorite database, your favorite message queue. All those different pieces that you need. Jenkins, et cetera. So what is that name? So DCOS, first and foremost, it's 100% open source. So that's really important. This is a project that we wanted to make sure everyone could participate in. We took everything that was needed in that box that I showed before, made sure all of it was open source. It's about 25 different components that we bundled together. And so by bundling those components together, we have a single roadmap for where DCOS as a project is going. We have a single set of documentation and tutorials and examples. Things that you can follow for helping you make your DCOS cluster run. Also, it's not limited in any way. So I think this was one of the concerns that a lot of folks had early on. Is this going to just be a community addition that's going to limit the way I can use DCOS? And the answer is no. You could use it on a vagrant in your laptop. You could use it on a 10,000 node cluster in production. It can run anywhere you run. And last but not least, it's familiar. This was important to us as well. When we picked the components that we wanted to put into DCOS, we were looking for components that people were familiar with. So it wasn't the obscure stuff. It's mesos, it's marathon. It's the things that you're used to using already. If I go back to the slide with all of those logos, it's all the technologies those guys are using just in a form that's much easier to install, much easier to actually take advantage of because it's a single distribution. So just a couple of screenshots because I wish I could do a demo, but screenshots will have to do. So I talked before about the CLI and the GUI. So we have a complete CLI for managing the data center. One of the cool things that I like about the CLI is it's extensible. So as you install new packages, you get new commands into the CLI. When I install Kafka, then I can run DCOS Kafka in all my Kafka commands. When I install Spark, I can do DCOS Spark and my Spark commands. So the CLI itself actually grows as you grow your data center with new packages and new frames. And then there's the GUI, of course. So if you want to go in and get a sense of what's happening inside your data center, be able to visualize the activity of your data center, you can do that through the GUI as well. You can use either interface for running your containers. You can use either interface for managing your packages. Then there's the universe. I just love this feature because it makes it possible for you not only to get the packages that you want up and running very easily, but it also encourages you to try new packages, try new things. When it's literally as easy as a single click, why not install a couple of different databases and try them out? Why not install a couple of different message cues and see which one is best for your environment or for the problems that your application is solving? We have 69 packages in the universe right now and that number continues to grow. So we're excited to see packages coming from all over the world and getting put into the universe. It's hard to believe, but it's actually been six months since we launched DCOS. It was on April 19th. It's almost seven months now. So a couple of quick notes about the growth of DCOS. First and foremost, what we cared about in these first six months was making sure that users could take advantage of DCOS and we've had more than 31,000 clusters created using DCOS as the underlying technology. That's huge. It's much bigger than we were expecting when we launched the community. And the reason for that is because of that ease of use. I want mesos, I want marathon. If I can get them in a distribution that gives me that plus all the other pieces that I care about, then I'm not wasting time getting those things running. I'm focusing my energy on the places where my business sees real value, right? So 31,000 clusters created. We have a very large Slack community. We have a very large mailing list community and active mailing list community. Now we heard at the meetup in Beijing that we also need a WeChat community. We are adding that, so that will be coming very soon. And I want to encourage you to come and join us in these channels and also help us create the new channels that are right for your community. We're not going to limit where the communication can happen. And so as we see the community grow here, don't be shy about making recommendations to us for how we can do a better job of connecting to you all and making sure that you're part of the conversation that's happening within the community. So we'll be adding additional channels focused on you, focused on the Chinese market. Also, in six months, we've shipped two complete versions and we have a third version, 1.9, which is targeted to go into early access at the end of this year. So we're also really actively working right now on building those features with the community. We're already starting to think about and talk about what would be in 1.10 and beyond what that roadmap looks like. So we're excited to bring folks in and start those conversations. We formed a PMC, we have working groups up and running. So I want to encourage you to come and feel like this is a community that you can participate in and you can get some value out of. And I'll talk about how in just a second. Ben talked this morning about the MESOS survey that we did and he showed some statistics about the use of MESOS, which is great, I'm super happy with the growth of MESOS and the technology. But there was one other stat in that survey that I found to be particularly interesting. When we asked the new users of MESOS how they were actually using MESOS for the first time, were they just downloading it and getting it up and running on their own or separately? 87% of them said no. The first time I used MESOS, I used it with DCOS. Now why is that? That's because using MESOS with DCOS is much easier for most of those users. For most of them, it's easier if they get MESOS plus marathon plus these other components all together because they're gonna have to install those other things anyway. So I'm really hopeful that one of the things you'll take away from here is that if you're interested in MESOS, if you're interested in using MESOS and these other related technologies, that you'll think about DCOS as sort of the new entry level into using MESOS. It's a great way to get started with MESOS and marathon at these other technologies and you're not alone. Just about everybody now that tries MESOS for the first time is trying it as part of a DCOS bundle. Okay, a couple quick examples. I'm gonna get to a little bit more thinking about what these other packages are in just a second but I wanna also give a couple examples because this will help, I think, solidify why people are using DCOS and maybe even a little bit of why MESOS. So the first is Autodesk, a pretty well-known company. They created a blog post that really outlined their experiences running DCOS. And two numbers that I think are interesting on here. First is the reduction of AWS. So their data center got much more efficient by using DCOS. That's great. And what that enabled them to do then was to reduce their total resources that were focused on managing and maintaining that data center down to a single resource. Now, here's the important piece though. That was not a cost-cutting advantage for them. They were not trying to cut costs. What they were trying to do was make sure that as many people as possible were focused on building out the parts of their data center that were differentiated, that were competitive. So when they didn't have to spend time managing their cluster, when they didn't have to spend time building and rolling their own set of components, that gave them so much more time to go focus on building out their containers and their microservices and their applications. And that's the real takeaway from this case study. It's an opportunity for them to be better at investing their resources into the things that differentiate them as a business. They're not alone. We have case studies up on DCOS.io and on mesosphere.com where you can see examples like this of how companies are seeing benefits from using the open source technology. And for all the engineers and geeks in the room, I have a demo that I really recommend you go take a look at, especially if you're interested in big data or fast data. This demo is from a company called Esri. They have geospatial data. I don't know if that translates or not. They have a bunch of information about location. So it's all location-based information. So for instance, they'll put a location tracker on a bunch of ships as they come in and out of ports in Europe. And then they'll show you a map like this, which you can barely see. But they'll show you a map that shows you how those ships are coming in and out of a port. And this gives you information about what the most traveled lanes are within that port. They run this system using DCOS, using a fast data stack, including Spark, including Kafka. And when you go out to this demo, you can actually see how they build that out in their data center. So I highly recommend it. It just gives you a sense of another kind of use case. We see fast data and big data as a very common use case in the market. And so this will give you a sense of how one company is actually deploying that. Okay, so now I've talked a little bit about what DCOS is. I think it's important. And I talked a little bit about sort of how we see modern applications, what that architecture looks like. Now I wanna talk about how we made the decisions that we made for which components we put into DCOS. So why did we choose Mary John or why did we choose Minuteman, for instance? So I'm gonna talk for a little bit about the characteristics of projects that we see being compatible with Mesos. And that'll give you some sense of why we chose certain projects over other projects. And there's three main characteristics that are important. First is a well-defined set of roles. One of the best things about Mesos is how focused it is on solving a very specific problem. Resource management. It does that really, really well. And what it doesn't do is anything else. And that's great. We want to see projects that have that kind of focus on being able to solve a single problem and solve it really well. So when we look at Marathon, for instance, as an example, Marathon is a great container orchestrator and that's what it does. That makes it a great fit with Mesos because they're both solving a very well-defined problem. There are other container orchestrators that also can do as good a job orchestrating containers. But they do a lot of other stuff too. It makes it difficult to be able to combine these technologies in the right way when the technologies themselves are trying to do too much, if that makes sense. So like a basketball team, we don't have a basketball team that's all point guards. We don't have a basketball team that's all shooting guards or centers. We have a basketball team that has very specific roles and that helps us do a better job of functioning as a team. And there's not one person that's trying to be all those different things. I had to put Kevin Durant in here now that he's a lawyer. So we look for that kind of teamwork within the data center. So that's why Mesos and Marathon are such a good pairing, I think. And that's why the other technologies, those 20 technologies that I was talking about before fit so well with Mesos. The second thing, modern governance, this is really important. And I think Steve Jobs had it exactly right. We look for technologies where the work of the community to build that technology is entirely based on making the right technical decisions. Where it's not a political decision, it's where we bring together the best and the brightest around that specific topic and we put them into a group where they can help solve problems together. Again, I look at the technologies that are in DCOS, most of them, a lot of them, are in Apache. Why? Because we believe in the Apache way. We believe in that model for helping to create great software projects. We think it works with Mesos, it certainly works with the other technologies in DCOS that are part of Apache and we have adopted the Apache way as the way we run the DCOS project as well. So there are some projects that are open source because they have an open source license. But are they really operating in a way that allows the community to come in and participate and engage and make changes? Not all of them are. And I think that's one of the things that I've always been impressed with Mesos as a project and that's one of the things that we require, we look for in other projects that are part of DCOS. When you think about building out your data center, you should be looking for projects that follow that Apache way or have that as their sort of guiding principle. And last but not least, an active user community. And I use this picture because if it was just people sitting in a design room designing an F1 car, it would be one thing. But they actually spend a lot of time listening to the driver. And that's really important because guess what? The driver is the person that has some of the most immediate information that you need. You can't just get everything from diagnostics. You have to be able to listen to the driver as well. They're the ones that are actually using the car. And I think the same thing is true with software. We look for projects that have an active user community because the voice of the user is super important. If you look at Mesos, if you look at Marathon and other projects around it, now if you look at DCOS as an open source project, we spend a lot of time talking to users. We spend a lot of time making sure the voice of the user is heard within the community. So over the next two days, you're gonna hear from a lot of users who are using Mesos and using DCOS. If you come and look at the activity in our Slack channel, the activity on our mailing list, there's a surprising number of users that are engaged in this conversation. So when we think about projects that fit well with Mesos, we think about projects that have an active user community and have users that want to contribute back. Remember, not all contributions are code. We don't look for just code contributions. We look for contributors that are also helping us define the roadmap, figure out the requirements, figure out what should be in the specifications. That's just as important to us as the actual code contribution. And so we get to the really important question. What do you want to build? I think this is one of the most fun parts of the conversation for me and I hope I'll be able to talk with some of you after the presentation about what it is that you're trying to build, how you're trying to take advantage of these technologies, how you see the different technologies working together, how DCOS and Mesos can help you solve those problems. So don't be shy, talk to me, talk to the other folks at Mesosphere out of our booth. We really want to hear what it is that you're doing and how you're doing it. And I'll just wrap up with a few sort of next steps. DCOS.io is the place to go to learn about DCOS as a project. It's where you can get more information and actually all the instructions you need for installing DCOS in your favorite cloud or in your on-prem environment. Everything you need is on DCOS.io. Unfortunately, most of it, almost all of it, is in English. But this is one of the areas where we could use the community's help. We would love to work with some of you to help us make this site more friendly to a Chinese environment. So if you're interested, if you go to DCOS.io, if you see something that you can help, where you can help us make this site better for this market, guess what, the site is open source too. The documentation is open source. It's all available through GitHub. And we would love to work with you to help make this site and the documentation and everything around it better. So don't be shy. If you'd like to help us out, we would really appreciate it. And we ask this not just of you, we ask this of every community member to help us make this better. As I said before, you can go to the GitHub, you can go to the Slack channel, you can use the mailing list. We use JIRA for tracking issues as well so you can go out to the JIRA and you can get a link to that from DCOS.io where you can see all the issues that we're working on together as a community. I'll give it a second because people are taking pictures. I'll make these slides available afterwards by the way so I know the resolution is pretty bad so. All right, there's another quick rundown of all the places you can find us and we expect we chat soon. So with that, I wanna thank you all for joining me in this presentation today. As I said, I'm not gonna take questions right now unfortunately because we're running a little late but I'll be available all day today so if you have any questions, don't be shy. Thank you.