 So good morning. Hey, I Can see everybody's awake this morning. That's good. So I'm a eerie ex Garnier I manage products for mesosphere and even though usually people already got it after a few seconds I have a French accent. I do live in the US. I'm based in San Francisco in the headquarter and Very happy to be here with you today. So yesterday was a pretty packed day and I think I really got I think during the main keynote in the morning Like I think like what a lot of people do which is at some point. I took my phone out And actually took a picture of the Netflix Katarina I will not try her last name. It's like nobody tried my first name So I took a picture of this because that was pretty impressive when she was putting all the dependencies She had between all the macro services the 380 I was even say services that were built at Netflix and how they scale that to millions of Users how they have like once again type of response time Monitoring of all of the service was really impressive and I was even more impressed because actually when you think about Where we are into that in modern data-driven applications all the monolithic apps, you know Billing it one single apps is kind of gone nobody's doing this anymore And I think you see somewhere some glance Where is that slide where you see exactly the same dependencies that we want to highlight here? Which is really about hey how we get all of this together how as an organizations you have to actually manage those Modern data-driven applications. How do you put to your containerized? Applications how you put your data services together and that's really I think one of the key challenges We see today around those modern cloud native application web scaling those being able to actually put all the rights Monitoring all the right process behind it and really evolving your infrastructure and your it to support those type of applications so when you think about that and kind of Making it like very simple and really in some way dumbing down when you think about your success You have to actually get two Really different type of applications and services running together one of them is the microservices or the stateless container which actually running but you do need with layers of more of big data services of actual, you know Analytics databases that you need to get to be able to actually Sometimes stateless is great, but you need to have some data in the back And you need to be able to persist some of that informations So putting those two together is actually one of the challenges not just about scaling But also making that run together and that's even more a challenge When you think about that explosion of all the different analytics engine Data stores you can see like a lot of names here But those apps which is coming out from the open source community from different vendors Which really are growing and you see the adoption you see the value and you have to take those Applications you have to onboard them you have to be able to support them because your business rely on them If you don't and if you think back at the Netflix Story why Netflix is successful because they have an amazing service because they have that service is back up by those Different infrastructure and applications and you need to do the same for your own businesses So all that explosion all of those is really coming to kind of a A choice for different infrastructure. You have a choice between control and some time Speed so when you think about that When you do those two approaches there is two school for that one is really going and saying hey I have my own data center. I'm actually going to be putting everything in my data center I want a full control. I want to know everything I want to be able to really like siloed some of those different workloads. I want to own it myself It's great. Give you the control, but it's also some time make it slower siloed Not really be able to react to those new technology not be able to be able to support them and so on so there is some value and There is some I would say less More inconvenient into putting this together The second approach is what we see a lot is a I can't do it or I'm a business I need a specific service my IT doesn't provide it. I'm just gonna go get it from the cloud Very easy. I can go get the speed the agility. I can go actually just get that service from Cloud provider being AWS Azure GCP or any other cloud provider But in that case, I'm kind of losing some of the control but more importantly You actually have some look into that provider because if I use a services Maybe I'm not able to migrate. Maybe I'm not gonna have that latest and greatest Which actually want to have so that looking can become kind of an issue and on top of that the cost even though it's kind of a Lot of time attractive when you start to really do the cost model and compare how much is your cloud Provider to how much you would do and pay owning it your own infrastructure Or even just using some of the basic resources from those cloud providers. You can see some difference within cost so out of those two approaches the kind of Approach that we want to bring is a third one is a different approach that is really the measure the vision of Mesos which is I want to bring together the best of both I want to be able to go and leveraging Mesos as That resource management being able to run all the different type of frameworks Being container orchestrations being data services I want to run those so I can actually have my Applications again those macro services those data services running on top of it So that's really that approach where you think about all the different agents which is running which could be running on any type of Infrastructure and that's what we really want to bring With leveraging the Apache Mesos, but also what we want to bring when we think about DCOS and some of you know heard about DCOS, it's actually a Open-source solutions that we kind of put together Which is really taking at the heart the resource orchestration of Mesos and being able to Expand on top of this what we did is put together all the different services Required to address that vision not only on the resource orchestration, but the whole vision about having a cloud-like services and with package it together on the With Mesosphere DCOS, so it's really being able to take and add Value like service deco discovery load balancing secret management all the different components you required in a much easier way Because it's packaged together It's easier to install either to deploy and again This is what's running on all the different type of infrastructure being physical Cloud provider virtual whatever you want being your infrastructure, and that's really what we want to bring in that type of packaging which is DCOS, so that's what we have today and One of the key Differentiations is actually the way we've built this if I go back to that more Synthetic view about DCOS and about that vision today The big differentiations is actually having what we call the application level or some people call it the two level scheduling It's really about how we want to differentiate What is that resource scheduler which is Mesos and how we want to differentiate this with the different workloads Running on top of it. I want to make sure that that content orchestrations that data service orchestrations Would be able to run in parallel because again This is some of the value that you have to deliver you need to be able to put those services together And you still want to optimize the resources which has running underneath But that's one of the key value and one of the key differentiations that we bring with with DCOS And actually we see that most of you are already using it if you think about and we talked about that yesterday There's Mac stack the egg stack all of those different I was gonna say solutions that we put together because you can't just run spark on the side You need to have like some Kafka you need to have Cassandra You need to have Mesos and so on when you put all of these together It's a much complex system and we see that running today on more than 50% of all the different DCOS clusters So it's already that vision is really the value from Mesos that we bring here and being able to make that available to you The second one is actually something which is a bit more subtle I would say for some of you which is really what we've added on top of this It's something that we've introduced maybe a year a bit more than that actually Which is called the SDK DCOS commons is like different names But it's really about hey when I want to build those different frameworks when I want to build those orchestrations How do I do it? Can I actually go and build it themselves? Do I need to put a lot of line of codes? How do I actually do build and maintain those different framework? DSDK is actually what we brought into the picture to make sure that you can build all those frameworks in a very rapid pace But also it's all the different management what we call the plans Which actually allow you to just define how not only you're gonna run that framework on top of Mesos and Within DCOS, but also how you're gonna manage it how you're gonna maintain is there the failure of a node How you want to restart or not that specific node how you want to restart the the survey is the task all of those Is what's actually built within the SDK that the different framework can take advantage of? It allows you to build faster allows us to build those framework faster But also when you want to manage them you actually manage all of them from always Single interface allowing you to have the same way of managing you want to update you want to restart You want to run the new services that will always be the same consistent way to actually do those Different management disregarding of the technology that you want to use and that's really one of the key value that that SDK is actually bringing It's that simple that we actually introduced a service catalog in In DCOS which allows you to have all of those service available. It's a one-click install I can go and pick one of those here like Kafka Elastic HD FX and install them in just one click That's something which we have in DCOS and we even have the notion of certified framework We have like higher level of certification will make sure that they actually run within each other and they run properly on top of DCOS so that's Really the two level or the app level scheduling the SDK Which make that differentiations and allow us to actually bring a lot of those framework If you remember few months a few weeks ago, I was gonna say at Mezos con in North America We actually introduced Kubernetes on top of DCOS and for us Kubernetes on top of DCOS is just one more framework We use the SDK we built on top of the SDK and it just allow us to just now give you a choice For the content orchestration. Do you want to use marathon? Do you want to use? Kubernetes yesterday you heard even Netflix have like Jarvis. They have different framework this is just a choice that you have now and we've been able to build this on top of the SDK and What we did is what we really strongly believe is the best way to run Kubernetes Out of the GKE because you have very easy operations. It's a one-click Install of Kubernetes. It's allow you to by default be high avaiable High available. It allows you to be actually secured. You have the monitoring the metrics You can have integration with the load balancing the ingress and so on which is all done in that one click deploy It's also hundred percent pure Kubernetes It's not like hey, we change it have our specific API and so on. This is pure Kubernetes Coming from upstream. We have it available and you can just leverage it We didn't change anything here So very easy for you to get your developers if they use it you can deploy run it for them in production and Finally, it's also again that value of being able to put containers data services United unified running on the same platform Leveraging the resources to their maximum by making sure that you have those running together on the same type of resources and that's really One of the value that is really key here. We're working that version of Kubernetes actually is still better for now We're working towards being able to g a that in the next few months and also being able in the future to make it run across Multiple versions of Kubernetes multiple cluster of Kubernetes. So it's even more value that we can bring on top of Kubernetes but that was a month's Like five or six weeks ago So today actually what we did yesterday. We actually did it again We're actually self-cuban it is we went to the next kind of very hot Project that we can see with getting a lot of stars, which is now I'm very happy to introduce that now We have tensorflow Running on DC OS. This is a pretty big achievement York is applauding. So if you want you can And that same thing we built it on top of the SDK. It's in some way was easy But we had very fast time to market because we have that and we be able to deliver it on top of DC OS what it means and for people who don't know what tensorflow it's that Open source software library for machine intelligence. It's not even my wording is like actually from tensorflow.org and you'll hear more from that from a cave in later on today Because it's going to do a lot of deep dive, but you can actually take tensorflow What does it mean is you know, I go I design my different models I use a primitive from tensorflow then I write my code and that code I kind of optimize it Most of the time to run on my single node on my laptop and then I train the code I start to continue like running this and I do what we call trend that model So it's model is more and more Intelligence machine learning deep learning. It's really learning throughout those different runs Pull it straight forward to do on a single server Now when you go to a more distributed tensorflow You go in a very different space the code you wrote for your single node You have to write it now for distributed node, which means I need to take tensorflow Push it to all my servers make sure it's available then that code I actually have to do the cluster specifications put the different IPs of the servers. I have actually the worker the parameter servers I have the the score all of those have to be Statically parameters in tensorflow because that's how tensorflow work today You have still to deploy the code push it and make sure it's still Training your different code doing this in the distributed mode is very challenging. It's very hard So what we do we again remove all of that we took the same approach We have the SDK we can include all of this within that framework that we've built for tensorflow Which means if there's a failure will we start the node if you need to push the code to those different Servers will do it for you if you want to actually not statically Parameter the different IPs of where the code needs to run. We're gonna do it dynamically. This is exactly the value We bring and there is Like a very unique way to do this and today it's very hard to do it at that very large scale And that's really what we bring with the distributed tensorflow So if you're interested we have a couple of sessions actually today where we're gonna go way more in-depth into this And show you how it works what we did where we are today This is still a beta version of tensorflow and how we're gonna be able to do this Toward yeah, and what we'll be building even more in the future behind that One of the key value and why we've been able to do it is also because when we think about machine learning a lot of Time you think about GPUs and this is something we've enabled in Mezos in this year s Few releases ago, so you know it's always that vision of hey how we can put some of the primitives Making sure we have the base available and then we build on top of it And that's really what we did here with tensorflow and also with Kubernetes before So that's really where we are. I also want to kind of go through What's next? Mainly around DCOS. What are we doing after those different releases? What we have and this is one of the key differentiations We're actually continuing investing in those data services. This is something really key for us We see again a lot of you using the those frameworks And that's where we're going to continue enhancing them bringing new frameworks So we're talking about tensorflow. There's nx net There is a lot of other framework out there that we also want to bring to you So you have that simplicity to actually deploy it. We also Gonna work on what we call the DCOS Storage services, which is a better way for us to actually manage all the different volumes and bring that Within DCOS as a like first-class type of citizen in DCOS We're gonna work and actually quite a lot around what we call the cluster operations backup, upgrade, install Restore when you do your backup. We're bringing also the notion of a full domain Avability zone to DCOS So that's really key not just because it's a lie you to place the different workloads But because that's the foundations we're building towards having multi-tenancy in the future and also having a habit cloud Solution when you'll be able to actually burst some of the workloads to different Avability zone or if you're using for example Cassandra You'd be able to define specific racks when you want to run your different Cassandra nodes So that's what we actually working on right now on On DCOS and I did forget the private catalog Which is something we have a lot of requests when you think about all those Different frameworks that we have out there Available through the service catalog which used to be called universe for the one we we know about that before and It doesn't really work into an air gap solution and that's actually something we're gonna fix or enhance more than fix in In the next release of DCOS So you have actually all that value and benefit in your own non connected type of environment so all of this is Really what we're bringing the key thing behind that is How we actually realize the vision of Mesosphere the vision of actually DCOS which is based on actually mesos and that vision is really clear It's about how can we provide a public cloud like services? Just like you get from a cloud provider, but from an open Partner ecosystem all of those different frameworks are open source DCOS is open source mesos Of course is an open source Apache foundations and we want to run this and deliver it on any infrastructure So you don't have to choose You can get control you can get speed you can get all the value that you're expecting And that you actually want to be able to deliver your business Thank you. If you have any questions you have my handler here