 Good afternoon guys, hurry. How's everybody doing? I hope Still have some energy left for you know things days coming to an end Before I start a quick just show of hands. How many guys have her know what quantum does? Okay, how many guys have played with quantum? Okay, not so much Yeah, because I was the quads title was quantum on quantum. So I want to make sure you know If how much people understand how what quantum does obviously the user story We're gonna go over how we use quantum in our internal production environments and how we run You know how network virtualization Simplifies networking so much that you can actually run You know another layer of abstraction on top of it like kind of like inception for networking if you will so I mean it introduced myself. I'm so McBear as a founding member of the quantum project I worked for a nice era till till about now Now at VMware we we are part of the network and security business unit We were actively involved with the open stack community We're gonna go to our user story of how we use open stack internally and what how we what are the cost benefits and agility? benefits we have Achieve doing so Excuse me So, yeah, follow me feel free to follow me on Twitter. I post about being a quantum how things are changing how things are evolving interesting user stories and and Let's get started. So This we had quite a few user stories today a lot of people don't know dove down to the technical detail the network Architecture or you know the high-level open stack deployment. I'm thinking a little slightly slightly different approach to it I'm kind of looking at what are the benefits mostly we achieved and We'll get a little into how we did it and you know, obviously can ask me questions On details of any of the ass any of the aspects. I'm going to hold off till then for Q&A so getting Started let's see. I'm gonna make sure I know what the time is We So like like all of you guys we were Nice show this pretty small company. We we create scale. We grew really rapidly Everything was physical or I mean virtual, but we're just virtualization no cloud kind of environment and that was definitely very rigid it it was not because You know developers everybody said I'll give me cloud. It's just that it was out of necessity we wanted to move fast and if you're stuck in the physical world it was difficult to move fast and That's what drove, you know was the main negus impetus to guess trying to figure out how we use open stack internally You know to enable us to actually accelerate What for what we do so our main use case was first You have a bunch of networking developers, you know, it's probably the best in the world So that that's a good thing But the problem is that you don't want them to You know put them on your production network we still run a network and If then they're experimenting we're building next-generation networking technology So we have a very different set of challenges than a traditional enterprise does because you know They can actually take your network down because we are doing things which nobody probably does So how do we actually isolate? Like it's essentially what we do for work for servers to isolate them. How do you do that for network? We have a product which does that so it's like how can we use it to actually self-enable us ourselves to move faster? We're a networking developer if you don't virtualize the network, you know, we actually will hurt ourselves So it was like a real necessity like this is going to open set clouds going to network virtualization Was the only way kind of out for us unless we want to take the rest scale, you know one day Some super smart crazy Networking developer is going to take the whole thing down just because he was trying to play with you know How we can how we can change how we do? multicast handling and In our virtualization layer or something next use case was We nice here. We're a very, you know fully software company where we make Distributed systems to handle network virtualization We deliver software really fast. We believe in agile methodology every four weeks for enterprise software That's you know, not unheard of and a big ball into achieving that kind of agility that kind of speed is continuous integration build continuous integration and and at low cost we were a small Scrappy startup. We didn't have couldn't dedicate hundreds of physical servers to do continuous build and integration testing So we wanted to have the agility or the scale of big the big guys You know, maybe we you know, I can how can we use our infrastructure of developer work work in environments, maybe at night to do continuous integration when nobody uses using it the only real Option is you know, you can't do anything unless you use cloud So like now we have a shared pool of capacity Developers so whoever is using in the morning the workloads actually die down in the evening You can use it for a different set of tasks You know for the throughout the whole night the same you're using a resource requirements by by a huge factor So that you know, so we have this dedicated build environment runs at that capacity at all the time And then when there's cloud capacity we burst into it Because we are using a virtualized network. There is you know, we can it seems as the same network and you just grab compute capacity They have L2 or L3 reach ability most of times we actually provide L2 reach ability and while out there you go You have more capacity when you need it and that accelerates our ability to deliver what we do best deliver software 30 use cases that we're a whole bunch of application. This is a complex You know exchange kind of application, but we have much bunch of applications like this our actually network virtualization solution is It's pretty complex and to deploy once we deploy it It's easy for users for open stack users or any of the cloud management platform users to virtualize the network But it is not easy to deploy it internally so The this use case was driven by you know accelerating. I guess time to revenue like we have to do proof of concepts Go to customer sites. How do you develop really complex applications fast? It used to take us? I'm going to go how long it used to take us but you used to take us few days to how do you bring up to the cloud and deploy it in in seconds we'd you know We did that by Essentially once you virtualize it we could author these labs or these applications and the topologies What what are the different elements which go different networks and kind of have this if you will manifest? Which goes with every one of them and then you know you click a button With within a minute you get the entire infrastructure deployed and we do do a whole lot of that because Every time it's a sales guy or overseas have to do a demo They need to do it and before you stick them at least a week to rig the physical hardware to do it and That was definitely inhibitor to our products at adoption and how fast we can actually demonstrate the value So that was this was another out of necessity requirement like how do we make this complex enterprise? applications work fast and and so that was our final use case and I'm and other few things I wanted to go over was Since we went down this path of bringing our own open stack cloud What are the and with network virtualizing network with quantum? Why did we do it and just give a contrasting? You know how it was before and what what were benefits we got out of it then we can Have any questions later on and any of this how we if you have questions on why we think we had these problems are How did we actually accelerate this? numbers I'm talking to you about so Before quantum I was talking about my first use case remember so Every developer pretty much we were a small company So we the only way we could actually make sure they don't bring down a corporate network was that every developer had had their own private VLAN and it nothing goes outside that boundary Anything else, you know, and that's also still risky. You know, these are these are pretty good networking telephones talking about We are networking guys are pretty nervous, but But even even with that isolation was not fast enough because people are moving between projects or they were to collaborate You know two guys two guys are working on a different feature. They had to you know, and you have to change their VLAN so they can actually They're collaborating when they're coding how do the test that if they're not in the same test environment, right? So every time you make changes it takes at least couple of days even for a fast Startup like us to actually get the networking guys involved that would change and give the right guys access to the correct VLAN everybody on the same page and That was a really big win for us to actually Me virtualize the network Click off a button you get what you want You have this notion of projects you can just share with multiple people you can add delete depending on you know Who's working on it? And you know, I say 20 seconds. It's really not 20 seconds It's like it's under a second API call 20 seconds to actually you know to drop drop down the UI select I want this network. This is these are the guys like I want to give access to this project There you go. So we got out of it, you know, nobody takes on the corporate network anymore. So So that was that was really big when it Helped us go the go down the path of this continuous integration fast development modules fast deployment second scenario I was going to talk about was that Like I was talking about our build and QA and continuous integration use cases Are as we do grew the product it became more complex more features that means means The oldest means one thing means, you know, QA is going to say oh, I need more resources to test it. I need more time Build stick longer And that that's how most companies actually slow down our most really large enterprise environments is just you go from, you know, TV development cycle to six weekends before you know six months and and everybody knows what happens after that and We from fundamentally we the company we structured we were all about agility Speed and flexibility. We're like we can't go down that path We can't have we have to fundamentally redesign the way we develop things the way we we have redesigned the way we don't do networking we have to redesign the way we function as a company and That that's how we we're like We can't scale by like incrementally adding more capacity or even renting capacity co-location facility none of them models work They're not cost prohibitive for the kind of scale we operate at we are you know our products Being used in really large customer environment really the world's largest telcos hosting providers Just was unfeasible. So what do we do? You know most networking guys. They don't give deck, you know Guaranteed bandwidth, you know if you need 10 gigs for application one day of the year you don't You know, you can't build you a lot of times you actually build the network to that capacity and that was not an option for us So we did what what is the next best thing we we try to? oversubscribe Or capacity and that and open stack or the cloud environment really led us to that and we actually started now Not only over subscribing, but efficiently sharing this as a single fungible pool of resources No A lot of everybody has their Macs or whatever where you write code But you need to deploy to some machine to test it to you know build build a few networks and run unit tests and such So developers come in, you know, they have their code wherever it's in the repo They pull it out they can launch up a VM to actually push down the unit tests to run the unit test Do functional validation And you know a lot of times most of people don't really need that VM anymore If if you have really good automation and infrastructure to do that and then so when they go home We there is no need to actually have that VM anymore. So the cultural change People have get used to it that everything has to be automated from ground up from day one Nothing can be manual can be that we can't do without automation and it's it's kind of a It's usually painful when you do when you introduce cultural change But it pays, you know, I guess many fold exponential dividends down the line because now everything is automated It gives you new I guess magnitudes of flexibility and return on your initial investment of that you did on automation So the developers we started trying to do encourage that any persistent data is showed on stored on volumes So, you know some kind of persistent storage by the VMs is End of the day they're done. We don't need them. You don't you release the resources So now all of the night if you see if we see our Monitoring graphs of our cloud, you know, it goes up and then it goes down and that we're like, oh now we can actually Multiply x that that capacity and I'll start using it for a different function And that's how we use it to accelerate our build and continuous integration by making sure our Compute capacity is always at like maximum utilization for different use cases and then we can prioritize who gets how much quota To make sure that we are always utilizing it the highest utilization. Usually can be capacity is really cheap. It's about most people have challenges is because it's really difficult to deploy to You know in a fast flexible way to use it to harness that capacity so that's what our open stack deployment kind of gave us and you know There was a result that we were able to continue to accelerate and innovate at the pace we were as a small company But having become a little larger with a lot more features a lot more developer We kept on moving at that that velocity that that business velocity is is kind of it's really priceless, right? I mean, it's not about the 20 boxes of servers you got but it's agility you you get as a business to actually grow and You know it achieve new levels which nobody in your Industry or none of your competitors could actually do. I mean so that was a fundamental differentiator or motivator for us to do this Next it was a my third use case. I was talking to you guys about those I was telling about you know every time we had we had To do a proof of concept our network virtualization solution We know it had these really humongous three of these boxes the particular set of you know administrative Networking configuration and just do some you know multi-destination traffic offload boxes Some device some other devices we interface with the external world. Somebody has to go install isos on them You know you have to get if it's a customer site or something Even worse you have to get their networking teams involved. They're like oh Who needs which firewall for these boxes on this network and it needs another another second network or you get switches and Yeah, it was just this nightmare. I mean I said three to five days But those are you know for really good customers who are all around here It took them were really smart and brilliant and have a very efficient networking and operations team it still took at least three to five days to get it up and running and While it was great, you know, we were when we were small We had a handful of people once we started scaling the organization There was no way we could we could keep up with that model like three to five days was just too expensive to dedicate and there has to be somebody to help you all that time right because Because most of time you're not doing anything fun or we know your sales engineers said they're not really Installing or demoing the product there and they're waiting for infrastructure to keep up with you know the speed people Want to proceed So now I'm gonna do a demo is we have the plus is a complex application later on It takes up about roughly about a minute sometimes less So that was it was not about going from three to five days to a minute I mean even though that that's a great achievement and there was a great accelerator now But it was about simplifying all of that now that we have codified that into a schema or Manifest I was like I was saying of this lab deployment topology needs these many VMs three to four different networks Firewall policies here and that's what I need and make it happen go cloud make it happen for me So that's what a cloud environment does you push a button you you get that right so and About you wait a couple a couple of more minutes. It's all up and running and That's that's that's humongous right it accelerates your time to revenue Many many folks like you know your sales engineering or human talent or human capital You have it's the most probably precious asset or in your company. So that's what we are actually getting out of it You know, it's not about money or saving on your capex or and you know All your op-ex or we are deacons infrastructure But the business value get a getting out of it actually going from you know five days to two minutes that's like I don't know exponential and the factor of Productivity increase you're gonna get out of your workforce. I mean and probably a good thing because most people when they do more They were going to feel more successful. They're gonna be able to do more. They're gonna be happy as a organization of the company so That was the wrong one This is supposed to be basically we increase your time to time to revenue, you know by exponentially many fold so Oh, this is oh, I went back. Okay. Yeah, we reduce your time to revenue by like by humongous Factor like Basiliant pretty much if you do the math of you know seven day or five days to do a single minute So At a high level that's that's why I see our chose open stack That's why we chose for network virtualization and at the end of it That's why he chose quantum with using network virtualization to enable a cloud environment That really helps us achieve our mission to our customers to our partners and be able to innovate at the rate We have innovated and continue doing so and show those show the world why we can do it and we'll continue doing so So the high-level, you know, this be many people who have been working with quantum Probably familiar with it. I just a high-level picture. I'll just talk to it how things work Now environment, it's pretty much how a typical quantum deployment works You know the user makes a request, you know through a nova controller use your ec2 apis or your nova apis You're like fire me a fire me up a VM We're using a sex for our current deployment. We're gonna upgrade to fold some very soon But right now we're using sx if you fire stuff a call to the cloud controller and the cloud controller Schedule gives it hands-over schedulers a scheduler puts on the message bus is like need this much capacity this flavor VM go get it One of the compute workers who has a capacity picks it up where would or who does this time for The compute worker actually picks up that request creates the virtual machine image and it has a networking Specs with it which are the networks in need to create three networks three then it creates actually three nicks for the VMs and In in sx we had some a notion called quantum manager. That's how we integrated it's changed with Folsom happy to discuss if later on if anybody wants to know how it's changed but So it essentially creates those next then makes calls to quantum using, you know The quantum or API saying I need this neck to be plugged into this network three next these three different networks and quantum go make it happen You can make it happen using, you know the open source open v-switch plug And we use And our product because we eat our own dark food and we had to make sure we keep continuing doing that so we use MVP the network virtualization platform to actually create a virtualized overlay network for every one of them and make sure that these nicks are plugged in and And you know then nova nova gets a htb to 200 request Succeeded command that means it's all it's good know what parts of the VM and no one really doesn't know anything The VM is automatically on the network. It sees IPs. It sees everybody all its peers So even this is virtualized networks from a nova perspective from an open-stat flow perspective nothing changes You know, it's the same thing you do with the open stack just that it's on a Virtualized network, so you don't have to get anybody involved. It doesn't use You know, it scales your physical network better. It doesn't expose all your max and In this in your cloud environment to the physical network So you build your physical network once and you can you can you know get You can scale it really better You can make you can have a lot more workloads Working on the same thing like cloud environments dynamic agile workloads where we are doing this every day I'll go with go with some numbers later on so We have right now is looking up cloud We have about two fifteen hundred or two thousand VMs running, but it's not about the these are not VMs We just spun up and we ran it We actually bring it up and down up and down every day these networks like you know Thousands of networks up and down and up and down every day imagine I mean your traditional IT infrastructure if you're a networking guy you have to be able to do that to actually Get the business velocity that you want you know to be able to For your software development for your you know customer side proof of concepts It just isn't fundamentally possible like that's that's like a bazillion like I was saying time correctivity increase It just cannot happen so And and then you know we get a VM Get a network So we make a few of those calls we have our typical Product is like takes about 13 VMs or 12 VMs four or five four networks You know these calls about take the total of our it's about a minute maybe two if if there is a lot of load and There you go and everything is done As far as details When a request comes in the quantum Frontend API it gets gets delegated to the plug-in you're using in a crisis NVV plug-in It goes to the NVV cluster the NVV clusters You know is a scale out just to be a system which actually manages all of the switching elements inside all of the Compute hypervisors we have we have a mixed environment actually we use KVM we use Zen because we work across everything So we our cloud actually has you know we try to support as many of them as possible in the single environment and And then from it like I already talked about the user workflow how things happen nobody sees a change under in the covers what's happening is the Network which led the controller is programming this virtual switches at every layer Just to tell it how to actually use the net you know Get a tunnel between where the whatever Whatever the packet needs to go so that's why the packets never go on a physical network They get encapsulated and sent it to the right destination Automagically somehow I'm happy to discuss more about that, but this is that's beyond the scope of this this presentation so So Once we had this I wanted to share is how you know the kind of adoption It's it's not that we are using a dev test some people say oh they're using dev test They're using production we use we are running essentially, you know if we're in we're a company We're enabling clouds are really large enterprise public clouds. We have to be you know We bet we're gonna better run our business on it So we kind of run across the across all sectors customer support training Dev and QA sales I see this is like real data. I got about a week ago We're just curious to work out the distribution we had so as you can see, you know It's essentially every part of a business runs and actually went to our top three, you know productivity boosts we got out of this effort actually doing We're running an open-sac cloud or running a cloud with using network virtualization, but it could cause each of these Verticals we had similar just unimaginable I guess gaining value and how much more proactive we have been so This is number of instances CPU utilization memory, but it's about the same you can say, you know the best most of the users are on Dev and QA and It used to be a Dev and QA and sales in SE, but we started kind of splitting You know this sales guys not kind of could do things faster. So this actually talks about How many hours people have spent also Kind of takes out into account So that's why the training and labs started taking more and more portions because that's where people spend time Before all of that was convergent was one single pie so So across all all aspects of the outcome company we have had Really significant productivity boost just because we could have get the agility we have never been able to get I would you know people with a lot of people who are talking about Does this thing work this table is open stack ready? So I just got a graph out of you know our adoption curve You know, it's it's in April weeks We launched this cloud because that's when a company started growing and we had to continue We had this immediate challenge and how do you do it? And this was the only way to do it was the fastest the path of least resistance fastest way to actually be able to solve Those problems and have that game that kind of agility. So as you can see we have a pretty steep Growth curve. It's pretty much mimics the radar company was growing I put a circle over there because that looks like you know, they had a failure. It was not really a failure That's in August end of August. That's when I see I was acquired by VMware So the IPs got the monitoring system got re IP'd so the This monitoring system which actually keeps track of it We use ganglia and it just lost connectivity. So it thought the world ended but It isn't what it looks like so that's But other than that it's been it's and the other thing we have seen as a step function That's because we kind of have quotas for developer It's it's it's been it's been kind of like a like a drug If you people because the more we give the more more the consumer and it just doesn't come and Another way to look at it out the way I look at it is It must it actually gives them such productivity boost that there's a lot of them Just can't go back to the old way of doing things like I cannot be that unproductive now that I've got in use of this fast Agile productive way of doing I'm delivering better. I'm performing better Every every person actually can do more with a lot less and only do more with a lot less But do things faster and do it himself without getting anybody involved And in our we are since we are we are really focused on networking We never get any of the IT network infrastructure everything is self-service and our network Infrastructure teams don't really care about it because it's contained. It's where it's virtualized. You can't do any damage You know you can you can shoot yourself on a foot, but you know be my guest do it yourself you pay the price Then this step function I was saying is because of quotas we have put in because of that at reason So as things things and I assume that we're gonna have some more hyperbys is a coat I will go up Other things to notice over here. This is the load. We are tracking The reason we are tracking loads not by number of VMs is because we I was like I was telling you earlier We try to fundamentally change the way we double software and the way we do business So it's we don't really you know Use the cloud but using the old way like spin up a VM and leave it on instead of my Machine under my desk now a VM in the cloud I mean that does give you a lot of benefits a lot of productivity boost but not Not in the way if you actually fundamentally change you are how you're using those VMs You're using as a femoral pieces out in there. I'll use it on demand I have everything is automated against you know You can use chef or you can use scripts or something else to deploy my application and get it going in a couple minutes Then everything is ephemeral so these are like load which we experience every day and at different times I've had to go a different graph. It will actually go down. It's it's it doesn't it's in a constant load So that's another big Benefit we Cain from going using the cloud is like the cultural change. That's actually the most invaluable Benefit of game. I mean, it's great clouds great. You get capex efficiency opex efficiency You can do things you couldn't do before but the cultural change organizational change That's usually the biggest barrier and once you actually successful implemented you have got that change You have a lot more agile organization. You can do a lot more things or You know invest resources and do it faster for your business priorities that you couldn't do before so So I don't think let's see how much time I have a little time so I'm gonna go and Talk about our demo topology what I'm going to deploy the lab house topology. I was talking about is essentially these We have Well, essentially I can go over what what it is, but it's a bunch of VMs need then each four different networks privates I isolated networks different security policies they can't talk to each other and That's because we do network virtualization some of these components are over the management network for management traffic and You know data network is essentially You know a physical fabric where you used to do overlays on top and you know, this is a demo topology You know, you know a sales guy goes and that is to show what's going on That's so I'm gonna show and a second sales guy comes in for a different customer He shows up another tenant so we can create a multiple of these and do it fast under within a minute or so So I thought for the do let's try to do the demo It's always a little challenging And doing one of these so So what I'm So here's our cloud UI. This is essentially a front end to Nova We have we could use horizon. We just had this use case of authoring labs There's nothing like that and we're moving fast So we just put an UI up against the Nova APIs and not doing anything fancy just using the Nova API just different UI So we have here is this thing called labs which are essentially You know a collection of manifest a bunch of VMs and networks and powerful policies How I want this environment to look like look like so I have this thing called open stack quantum training lab You can see the requirement needs like you know 40 meg gigs of RAM 12 CPUs seven machines Seven servers, so I'm gonna go ahead and oh So here I can monitor the status using This is because we are using the MVP plug-in for quantum and see there it is you know, ma'am It's an I'm the tenant. I have no switches. No ports I'm gonna go ahead and Deploy this lab And it probably take me a few minutes to Okay, so you see it's just starting to build out all the VMs It's an ad VM to show the external connectivity. There's the network controller. This is a this is a quantum demo lab I'm deploying actually We were actually gonna use this if you guys have time tomorrow morning to enhance on lab on using our cloud to Give it experience with how to deploy quantum and you know install it and get familiar with the commands and such So I'm deploying the same lab What we're gonna do for all of a whole bunch of as many of you as we can tomorrow to get experience with it so it's Doing it as we have seen in the networking side. We can go in debug and see What what has happened we can see that there are a whole bunch of ports which are these and Logical switches which are essentially our networks which have come up with this lab needed and They're live that it's gonna take a few minutes to spawn So while it's happening. This is happening in the background This is doing everything at traditional what we used to do in I was saying like five days, right? You rig up a server you give them this much capacity create the network and then there you go It's it's done creating the entire topology now the infrastructure is created essentially of these that complex app I was showing but The VMs are actually now installed, you know booting up and some of them are booting up from a remote volume and attaching them And there's scripts inside them to actually do that and that's what's going on in the background so I'll just go ahead and show you the network connectivity that all the network is in up and Show you that they're real VMs this little latency might take me a few minutes, but we might have to actually you guys might take a little more than a few minutes before they're actually all booted up fully and pingable and stuff, so I Will I guess I'm gonna start taking questions now you have and while this thing happens in the background and then Once it finishes I can do the ping test and such So if you have any question, please come on to the mic And yeah, we appreciate it. Could you talk a little bit more about the 10 lane is Maybe contrast it to open v-switch and gre and then maybe talk a little bit about the roadmap Are you gonna be extending to things like VXLan and nvg re? So darling, it's from an open-sac perspective. You don't never see a tunneling protocol You see I want a virtual network. You get a network I want these VMs on this network. You get it as far as what the nice US product supports. Yeah, it supports STD, it's an ITF draft. It supports it's gonna support VXLan. It supports gre and it can use any tunneling format It's just a low-level protocol detail of how you do virtualization But the real value is from an open stack from a user perspective The value remains the same like you're gonna get virtual networks and this is gonna work from day one Go ahead So it appears to me that you have packaged a bunch of VMs the network policies and the security policies together using some template I was just curious about how you created that. Is it a an open-source template? so So the template of this app I was showing the that's why so we created our own UI It's just an XML descriptor file saying that these are the components I need and based on that It makes a bunch of calls to Nova and quantum to get the network set up get the VM set up and plug them in just a Refinement if you would say on base open stack API's and because that was our use cases for in our pre-sales Kind of function. This is still booting up. So It's gonna be a few more minutes before actually Is that available While open source that template It's it's nothing really fancy. You all the API's are there you can Everywhere everybody's actually Free to take those API's and it's really easy to Cobble something together. We had some but a couple of guys who did it and did it in a free time over like three four days So are you doing? So said if you think there's any value and it would seem very specific to our use case I mean I see a lot of value in that are you doing firewall policies netting Whole lot of you're grouping everything together for the lab in that XML descriptor right pretty much what you need So this policies are actually There's one distinction to be made. They're just calls to quantum quantum extensions So we are not actually make calling firewalls. We're not creating firewalls or anybody doing not doing anything logic This is all front-end. So it's just making API calls So, you know anybody with you got a good Ruby on Rails developer and you give them the Nova and quantum API I don't think it would be really hard to we know turn it out of work Yeah, any more questions? Okay I'm trying to see if my other VMs are also booted up. It's going to take a few minutes But yeah, but I think it's going to take a few minutes to boot up You know, I'm running out of time probably one minute left So you guys are happy to come back to me at the VMware booth. I'm going to be there I can continue finishing up the demo and things and such Thanks very much You