 Good morning, I'm gonna check if the mic was on but I can definitely hear it is on Day three of ODS in Vancouver Introduction Chris Kenyon I look after sales and business development at canonical We're clearly gonna have quite an intimate session this morning So feel free to move a bit further forward. We can probably make this a really Yeah, I don't have to like that But feel free to move forward and we can probably answer go sort of do Q&A as we as we get going How's the show been going for everybody? Awesome, it's your foot. I'm interested how many people is this your first ODS? Oh Interesting like over half. What's the impression so far? Yeah What are you what which sessions have been most interesting for you so far and where are you from? Fantastic, well, hopefully we will answer all of your questions about how we work with the partner ecosystem I'm quick show of hands who here is using open stack in production right now in the room about half Which version is you on ice house anybody else? Havana Anybody else and then who here is using a bun to open stack as their base for their cloud About half the room more more than half the room. Okay, fantastic So what we're gonna talk today? What we're gonna talk about today is how We at canonical Make sure that a bunch who offers you the best possible ecosystem of both hardware and software and components For open stack and to do that. I'm gonna take us through This I've joined here with my colleague Jason Hobbs who is responsible for a remarkable project that we're gonna talk about Call the open stack into a property lab, which is at the heart of how we test and validate and guarantee You can build a cloud of your choice with components of your choice and yet to be confident that it's that it all works together That it's all gonna come together when you go to deployment so What I wanted to start with Was to talk about what our vision for what open stack should be for most users? Okay, we come to an event like Vancouver. There are there are always earlier doctors up on stage who Will be doing open stack on a hundred thousand a hundred and fifty thousand nodes often with our usual network topologies And it's quite a crafted open stack experience And that is great and these are the the super users are a fantastic For pushing open stack forward It's also true that there are many other organizations here who need to use open stack in a much more push-button Sort of way, you know in the end Infrastructure as a service is a commodity and you want to consume it as cheaply as possible So we're at the beginning of open stacks emergence and standard enterprise and it's good that everyone's here this week But frankly in two years time for really what you want from open stack is just that you press a button It works it maintains itself it upgrades when you want to upgrade it And really you want to have that to be as push-button and experience as possible Um from our perspective at canonical, you know all this if you look at the super users Comcast Walmart Who else did we have on stage yesterday? There was or two days ago? There was a digital film tree the good news is they're all building open stack with a month, okay? But we also want to make it this push-button experience and our the best way I can describe that experience of how well integrated all of this these partners and these these Ecosystem players should be in the open stack experience is to talk about Autopilot so Our autopilot experience is a Installer and management tool for open stack And I'm going to talk about it for just two minutes and then explain how this links into the partnership part is I can take The installer now it's in beta and I literally just name a region On a cloud and then I get to choose what we think are the most important decisions about how you build that cloud I can choose a hypervisor choose a networking stack object and block storage solution and Depending on the choices I've made the installer will say well you're gonna need if you want to do that in HA You're gonna need seven eight boxes If you make choices that we think would give you a cloud that couldn't be upgraded To the next version of open stack the installer will stop you it is it is intelligently guiding not just automating any decision It's saying these are good decisions to make for a an open-stack cloud that you can install manage and upgrade efficiently And what you'll see here is at the moment there's a limited number of choices there You'll see only KVM open v-switch chef or swift But the goal of our partner program is that you will have the ability to Choose a whole selection of components in there. So both KVM and ESX and Eventually hyper V in fact, we've seen a Microsoft partner demonstrate Ubuntu open stack based on a hyper V as a hypervisor in the last month A company called cloud base and so you have a choice of hypervisor for networking What we're going to demonstrate today is that all of the networking vendors we talk about all the SDN vendors They will be there as options in that installer So you could say I want to do contrail or I want to do new arch networks And that will just be a configuration button on that install experience and so on as we go to object storage That'll be obviously Seth and Swift, but you'll see more traditional vendors in the storage categories and be able to install a plug-in to VNX Viper copperhead from EMC. We announced a partnership yesterday with next center. So a whole wide array of partners that will be enabled in that installer And we think that for people running a cloud that's under maybe 300 physical hosts This is how you should be installing open stack by the end of this year It really should be this this this simpler process And that what that install will do is it not will not only do the install it will then help you manage that That open stack cloud going forward So that's the vision of open stack as a push button experience that you've got choice about what you do So we're not locking you to one very very limited build of open stack You've got choice, but we are guaranteeing that you're on a known good configuration That's pretty critically that is upgradeable. Why do we hark on about upgrade ability so much? How many here how many people in the room have done an upgrade of open stack? From one version of open stack to another on a production cloud on a production cloud How was that experience for you? hard Yeah, how was that experience for you? Sorry a work in progress. Okay, so some scars, but but okay Which version did you upgrade from I found out do you know which which versions were you upgrading? Okay a while ago That's how That's the vision for how to do it the reason we hark on about upgrade ability is because the key the killer functionality of open stack as you will have all Noticed from being here this week is that it's moving so quickly The best feature you will ever get from open stack is the ability to upgrade in place And so that's why we're so focused on that that upgrade story So as an install process does that look like something that would be interesting? You know if you wanted to experiment with open stack quickly or get it set up Would that be an experience you would want I'm getting seem to nod it nodding at the back of the room Okay, so how do we how do we get there? Underneath the hood of autopilot. We're using two tools One is mass metal as a service for doing bare metal provisioning and the other one is juju And juju is acting as a universal model for all of the services that are in open stack itself And here you can see for us a typical HA deployment of open stack, and we've got it running there with Nuage as an SDN And so what's critical here is for us to be able to deliver that one that that Guided install journey. We need partners to have taken their Solution and to charm it up into what we call a juju charm and Once we have that we then have the ability to do an automatic install with any partner Component so you can see that if we had different different architecture here. I'm swapping out nuage for an architecture where we're using calico, which is a layer 3 SDN from meta switch who also do the project clear water and All I'm doing there is that the architecture is staying exactly the same and I'm swapping out the SDN vendor So what's important for the partner is that they've taken that component They've charmed it up and then that gives us this ability to automatically build a cloud with their component installed so To To do that To give this ability to integrate lots of different solutions into open stack We need to address this hard problem of interoperability Okay, so the beautiful thing about open stack is it gives you lots of choice and With that choice comes this cost, which is can we deliver that flexibility with high quality? And the solution to that is something called the open stack interoperability lab that we set up over 18 months ago Dan correct me if I'm 18 months ago is about right In Boston and it is the largest interoperability testing facility for open stack in the world Okay, it's been going for 18 months. It's fully automated and What we're doing there is we are taking permutations of how you'd build open stacks so at its very base layer compute storage and networking and Starting to map out all the different permutations So someone here you may be doing KVN Seth and Cisco Nexus as an SDN But somebody else may be saying actually I want to build with ESX Netapp and Juniper as my options and you can see very quickly. We're going to start getting to quite a lot of permutations here When you start then layering on top of these hardware permit these these component permutations If you then start adding in Version permutations so going from ice house and Juno. We start getting into a really big automated testing exercise and that is the that is the work that happens in the open stack interoperability lab So just to give you an idea of where we are at the moment There are over 30 vendors who? Who pay to have their equipment in the interoperability lab? This is not a logo swapping exercise This is a how do we as an ecosystem? Guarantee to you as users out there that these things will work together How do we ensure that we spot? issues with tip between You know suddenly discover that Seth and Juniper on ESX are not working well together How do we make sure that information is going back to vendors in a timely fashion so they can fix those those issues? So just to hit you some numbers Over 73 hardware combinations are are in oil We have over a different over over 175 different combinations of how to build that cloud In terms of which services from which vendor? We build automatically over three and a half thousand open-stack clouds a month Okay, so hugely automatic largest testing exercise that's over 46,000 open-stack deployments And since we started the lab, it's over 1.3 million functional tests have been run So we're doing this at extraordinary scale If you're a customer one of the things we will always do in our first meeting with you is simply say You know one of the combinations that you are either have deployed or are thinking about deploying So a lot about this is then reaching out to new vendors and making sure they've got their equipment in the lab And that they've given us both the the solution that they have that needs to be tested But also functional tests around it So if you're Juniper contrail for example You give us functional tests on top of contrail that prove that contrail has correctly come up in his functioning and As part of the automated infrastructure in the test We will bring up that cloud and then start running your functional tests as a partner on top of that So maybe to talk a little bit more about how that happens in practice I'll hand over to Jason and you can maybe just talk through how the lab is set up. Sure the room, okay? So this is all based on the boot you up and stack, of course So we're deploying a booting up and open stack Have our core components along with our partners technologies whether those are hardware or software based or a combination of either So this is this is all completely automated at any given point in time We've got about 12 clouds either under deployment or under test and the way we do that is using Juju and mass so we use mass for managing that the bare metal layer And use juju to kind of coordinate what what mass is doing we've got Around 150 servers that we're deploying open stack onto so So I said about any good point time. We've got about 12 clouds being deployed or tested and and so the way it works is We've got a piece of software we call our our Q master that looks at all the possible combinations of open stack services and and picks a combination and submits it to We use Jenkins for running tests. So submits it to Jenkins and And that gets juju and mass kicked off and deploying open stack Once open stack is up and running And it's it's ready to use. We've got a second stage. We call prepare that Configures up in stack for use so it sets up networks add users uploads images to glance and gets it ready for for actually using and testing with And then after that we've got a testing stage where we run tempest We run any of the tests that partners have provided to run against their their technologies and so once all that stuff is done we've got a Kind of an analysis stage where we look through all the logs for the deployment. We look for for signs of failure so trace backs fell tests and things like that and Identify identify bugs we can classify bugs by whether or not we've seen them before So if we haven't seen a bug before we can flag it and have an engineer go and investigate the source of failure and Figure out what what went wrong? When we find bugs will report those to partners if they're bugs with partner technologies And then so all kind of all those results from each of those runs It gets we save them into into Swift. We keep them forever And on a on a monthly basis will go and pull all the results for for each partner and generate a report for the partner that says kind of you know, which which services Did they deploy with what combinations of they deploy with how many times I did deploy what kind of failures we're involved in their deployments And so we'll send those out on a monthly basis Test that's kind of how it works. It just can keep going constantly 24-7 We do about 150 deployments a day It's pretty cool 150 open stack clouds a day is anybody doing that many opens clouds a day internally So questions about how we're running that lab in the room I'm either partners or customers who got questions about what's happening in the lab and the tests that we're doing go ahead Yes, it's a simple answer. So So the question I don't if people hear it is does there have to be hardware in the lab to do that do the testing? So obviously some of the components in the lab of software, but but if it's a hardware component We absolutely need to have the the hardware permanently in the lab. And in fact, you touch on something I was going to jump to Which maybe I'll just jump to So I'm doing this a little bit blind You touch on a topic which Here we are which is the difference between third-world hardware certification for canonical and being in the lab so Here we're talking about continual integration of all of these components with each other and it's for us It's a separate thing from hardware certification Hardware certification is typically a relatively one-off Part of the process and actually I thought I'd bring this up on the screen There we are. So that is looking at a matrix of certified hardware For Ubuntu itself. So this is not in the lab This is looking at Ubuntu itself and there you can see from HP that you've got over 41 servers certified with 14 oh 04 last LTS and 64 servers certified with 1204 and you can see that a subset of them only nine are actually in the lab Okay, so to be in the lab We need to have the hardware because we are testing every day. We're also now Starting to pull down the daily build of open stack next release tip and testing that here So we're getting we're giving the vendors very early warning that there's an issue with a commit That's going to break some of their compatibility So the answer here is if you want to be an oil and you want to be part of that program you absolutely have to have that hardware in the lab permanently and Yeah, this is simple answer That's absolutely right. So here on if I look at this particular ProLiant device, it's It's certified with Ubuntu. So that means You got to remember at Canonical. We're we're also gonna we're operating system company as well as open stack There are 50 people in Taiwan. We've got employees in bed at a HP Cisco and they'll badged on-site Every time they come out with a new piece of hardware in fact pre-production hardware. We're certifying that with Ubuntu itself but a one-off certification of a Server is different from continually testing it with all of the other components And so it is a different program and and the vendors, you know the partners Have to engage very proactively in it in a sense having your hardware and oil is interesting But not enough what you actually also have to do is then be willing to take the feedback every month detailed feedback and actually start You know getting through the issues that we're discovering the bugs that are being discovered Which then which company you from it's actually okay And I know there I know there are some conversations going on going on that Perfect question. So three things. So firstly is engaged with with us directly So we have a dedicated team that run oil from a business development perspective So make contact with them. What we're looking from for you is hardware availability test harnesses that are specific to your Cinder plug-in or your particular hardware There is a financial commitment to actually help run the lab So this this lab is funded by all of the vendors that are in it You know conical conical makes a contribution, but most of the the Funding for it comes from vendors and that's what you need to do So we happily talked to you after this session about how to get that in there So does the difference between hardware certification and oil participation makes sense to people in the room we That's clear. Okay so Let's talk a little bit about what you get then as a vendor out of that So you will get a sort of 30 40 page monthly report from us. You get live, you know, we will we'll give you all of the bugs In launch pad so that you can start dealing with issues live or you can wait for the consolidated reports Sometimes the consolidated reports add more insight I mean a lot of stuff here fails and so some of it's about statistically tracking whether or not a failure rate is Significant or not. It's not purely just a green-red on functional testing Sometimes it's it's analysis of overall whether or not the the failure rate is what we would expect or not once So to give you an idea of people who in that that's a present subset of partners Sorry, that's a present list of partners who in in oil And it's everyone from compute networking storage increasingly. We're now also starting to see vendors who've got plugins to open stack which are not core open stack But companies like Talagent for example have a really great billing system where that plugs into Solometer who also want to get test feedback on How their components is behaving across multiple the different architected clouds and we're seeing them come in there So any questions on current partners? Is there any anybody who's using the bun to open stack out in the field who says oh my word? I didn't realize that my my main supplier is not on that list Any questions on that list? So partner reports Obviously, there's a detailed report. I pull. This is just a summary page HP allowed me to share this but it this is the This is a sort of report you'd get back. Obviously the number of clouds were tested on which configurations Which specific open stack services were testing which aligns pretty well with dev core And then also we are testing both red hat oracle and the bun to guests on top of those clouds It is unusual for the guests to throw up a particular problem with a cloud, but we are doing that testing In case that in case that it does and you can see here that for this particular report Which versions they're getting feedback on and also obviously which services are being tested in that environment. So very structured Very structured setup So in a sense, that's what you care about but if it turned out there was a cinder key I mean, I can't imagine why there would be a cinder keystone. We're giving you the full We're giving the full data. It may well be that in the report. You you just assume that Anything that's error outside of cinder is not your problem or unlikely to be your problem and we focus it on just the cinder issues but We give you the full the full suite great question. So up until now it has been What do you want me to answer this? Yeah, sure. We're testing out of the blue to cloud archive So so we start testing you open stack releases before they come out. So we started testing Keto about a month before it came out and And then as it comes out and gets released into our where we package a stable Open stack releases we test out of that So it's the same stuff that you would you would pick up if you're installing as a user installing to bring to open stack I'm gonna do a little demo in a second. Give me a second Any other questions in the room while I just get this a little demo? So the question so the question is what are we doing for server certification with someone like HP? Yes, yeah, we've got a kind of a standard so we did a test that we run There's there's tests that run on the server So so we have to in order to certify we've got to get one of your servers or sometimes we'll do remote testing or Go on site, but usually we'll get more servers in our lab And we've got sort of a standard test suite the checkings like you know the storage work does the network work? Can we install does that work? We also test with with mass so whether or not mass can Can control the system? So So that that kind of certification process and clears the way to get into oil if your server gets certified Then we know it's gonna have a good chance of working well in oil Given we use mass to control everything in oil So I can't we can't hate you So the question is as a server vendor you have choices of storage adapters or network adapters in your in your servers So I think typically and I'm not an expert on certification I think typically what we want since there might be differences in functionality with those different storage controllers Or it might work with one and might not work with the other we want to have separate certifications for each of those You know each of those different models that has different components in it like that Yeah, so so there's a dedicated hardware sort of server certification team They'll have a matrix of things they do and don't care about obviously a bit of extra RAM is unlikely to make a big difference But maybe a different storage controller would and they will work with so which company you from? Sorry Qlogic so that the certification team will work with you and decide which variations they would want to see a new piece of hardware for In the certification lab and which ones they'll just accept are statistically unlikely to be an issue That is a pretty that's a well-oiled machine that the hardware certification teams have been going for so six seven years And there's probably 60 people in it And you would your team be operating out of the US or out of Taipei? Okay, so we can that there's hardware certification Leads in both both markets and we would hook you up with them. So we happily talk after this session Okay, I wanted to just do a live demo In a second so I got that set up So where does this get us to so you now know you get you've got confidence that you can build a cloud of these different Components well, you know something that a lot of you might be doing who's here is evaluating an SDN this year Anybody looking at SDN evaluation this year or making a decision about SDN People are holding off until next year. Is that right? Okay, large a lot of our large customers right now It's starting to make decisions about SDN to do that you really need to have it connected up to a cloud and Doing the SDN evaluation might take you a week or two easily to get set up So what you're now able to do by taking the the charms that we're using for these automated builds in open stack is you can now get from the vendors those charms and we can give you Validated known architectures of a high an HA open stack cloud with any SDN vendor and you're literally able to drag and drop The the service description YAML file from juju onto a new cloud bare metal Rack of of machines and have that clilt that That's an architecture built automatically for you And that could be new arch. It can be calico. It could be this is Cisco who funded the work around an open daylight integration with Ubuntu open stack or it could be Metacora plumb grid, there's a whole set of SDN vendors all coming in in the next four or five months into the lab and in fact we go live here. I'll bring up With me a second I'll bring that up On screen Right now here on the I've got in front of us two orange boxes so Here there are two orange boxes Which for those who don't know the orange boxes. They've got ten Intel I fives in them with shared storage and a switch So it's a data center in a box and here on this side. What we've got is juniper contrail installed on that box And I am able to take That's that's running on that device You should have ten lights on the front showing that all of those boxes are on and in fact I can take that architecture and if I wanted to share that with somebody else in the organization I can export that I then get a simple YAML file which describes that particular architecture Let me make that a little bit bigger And that is describing in a YAML file all of those services And I'm able to then give that to a colleague with all the relationships set up And I can literally just go and build that on a fresh install. So it's completely Changing the speed at which you can experiment with different SDNs and we've got known good bundles For all the vendors I showed with another four or five coming in in the next couple of months So if you're doing an SDN evaluation project, this is the easiest way to be All right, so I'm onto a Onto the other orange box where I want to do the install of The install using the autopilot so putting this all together so Okay, so let's have a little look so on this on orange box on the left from where you're sitting You will see that we have Hopefully one What you'll see here if you can see at the back is we've got Metal as a service looking into this box and it can see 10 physical nodes and you can see that only one of them is deployed Which is this physical node here and that should correspond on the front of this little box So that one light that you see here So one physical node and it's got the installer app that we're going to look at right now Installed the other nine nodes are completely blank right now. There's nothing on them. No operating system and Now what we want to do is we want to build our cloud so I'm going to log in to that node that has the installer and Actually, who here has not installed open stack before You install open stack before You have somebody who hasn't done open stack before come on hands up. Don't be shy I'm going to pick on you to the front. You can join a couple stage. Yeah, come up here. We want to build a cloud I'm gonna do about a build a cloud live Two or three weeks, okay, let's see if we can do that a little bit faster because I know that people have got coffee to get to In about ten minutes. Okay, so we need to give this cloud a name so feel feel free just to give it a name for the region What is your name, sorry? Where are you normally based? Fantastic, that's good. That's a good name and we sort of do want to call this we can call it cloud one or you can call it something else Okay, so we want to choose a hypervisor. So if you just go down you'll need to use a pointer Yep, choose KVM It's pretty simple. So just select We've only got open V switch available But soon this will have Juniper com trail and new arch networks will be embedded in that installer. I Need to put a gateway network in there. Don't we Doug What And it's in the public gateway there Believe we have that there we are. Okay, so I've just added up the Forgetting out on the public network. So select the rest of it. Yep select That's also populated by having discovered. What's on the device? Or is it in the same network? I think you got the yeah, 27 so 72 Typo Can you get that there we are 27 And a dot I think you lost Okay, so that's some go down and select Me to help you on the Sorry, it's the the think pad revenge of the think pad pointer. There we go Okay, and I will select Thank you. Please. There we are. You should dream that. Okay then Seth and Or swift as you see fit. We're just gonna hot add hardware So I'm gonna just and do this quickly because we've got people coming in There we are and we will need to add a little bit more Hardware and we didn't make That's it and then install and that Is now kicking off and for people who are staying for the second session where we're going to talk about How Lexi crushes KVM you will See on that left-hand box that the lights will start wearing and by for the end of the next session finishes You will have a full open-stat cloud that you've just done live on stage. So thank you turn on Now I should take about 20 minutes. Thank you very much. Okay, so that's us wrapped up. Thank you very much for joining us on