 All right. Good afternoon everyone. Thank you for joining us to the taking open stack to the enterprise session I know it's pretty much the last session today. So you are actually Can give yourself a good note of surviving the whole day with all the session. I know it's pretty hard to absorb Open stacks on the sessions and I think the topic we have today in sake is very important one, right? It's called taking open stack to the enterprise and what we'll try to do in the next About 40 minutes is to understand first of all what's enterprise mean, right? Because we heard a lot of buzzword enterprise sessions today around that name, but we want to understand what actually means and and to zoom into What we have done in order to deliver a solution that we believe is good enough for Production in enterprise, right? But that's the missing piece probably that we're missing in the enterprise ready With me after today our cutting I pretty want to Present yourself. I just want to my name is Sean coin I'm a principal for the manager for open stack in red up and I'm gonna cover the our side of the story and I'm Arkadiy. Can I ask I'm a director of development on the delt side jointly working with Red Hat team on delivering you the enterprise level joint solution? Mr. Kroche who's our PM who's supposed to be here unfortunately have to live for the family emergency So I'm kind of stepping in to take care of his portion of the presentation Thank you, Arkadiy. So we'll stop by looking at the evolution of enterprise in open stack As I said, what is exactly who is exactly the enterprise that we split speaking about? And why is opensack good for enterprise and vice versa, right? The notion of open stack to be a scale out Maybe just maybe for public provider, etc. Is one but we also have a very large interest in enterprise and we will take a step at that and Then we're gonna actually zoom in to our approach and why does it matter? In the one sentence we have released already the third version lately of the cloud solution that is based on the joint solution of reddit and delt and It's the reference architect that will have for you to as a link at the end is out there But we call it a reference architect, but it's actually much more than reference architect, right? So a reference actor is is it's good for reference from limitation But it we actually offering a whole solution that you can actually stand out tomorrow morning that we know Works and that's the key, right? So open stack is a very complex Infrastructure to set up and deployment is one of the key aspects that will touch the And then we're gonna talk about where we where we're going What's what's what's the future looks like as well some of the key takeaways? so we'll start with just looking at enterprise mentioned in open-stack summit and When we look back at Austin Diablo Pretty much then with that. We don't know and we don't care about enterprise, right? We just starting the project and There was no even notion of enterprise Features or whatever enterprises ready or not. We just wanted to stand up a cloud infrastructure Software, which is open source and stand it up, right later on moving to Essex in Falsum We started to gather requirements and how the enterprise can actually benefit from it and It's getting ready for it, but not not really only in Grizzly in high sales We moved to what we called stabilizing mode when the summit talk started to appear with specific vendors Enabling open-stack for the apple sack. So if I had any beckons supported if it's a storage or network Story I can connect it into open-stack and make it enterprise, right? so that was pretty much the notion and then moving up the stack starting to look at Juno and keel release We actually starting to have discussion around the core is ready and Gartner actually took a stab at it and made a statement that Juno Release is the enterprise ready 1.0 release, right and the reason is a lot of the key features That requires enterprise adoption only made it in into kilo release and we'll take a zoom into some of highlighting some of them, but before we do let's take a notion of what's enterprise it and typically when when you look at a Solution promotion data center solution, you will see guys in suits standing in data centers Great, this is that but guess what? We're not open-stack as I mentioned earlier is a very complex Infrastructure to stand up This is not we're referring as an enterprise IT also not as consumers because if you look at the persona that actually consume and stand up Cloud services today is it's not the classic old-fashioned when the only consumed services today IT role has changed to be a service provider within the organization So if I'm a consuming or standing up a cloud infrastructure, I need to actually provide Services within my organization or organization that include different service catalog items can be storage compute, etc But in a cloud fashion, so in a sense, this is no longer representative or enterprise IT and when we look at what enterprise IT and who they are so So would there are I know it when I see it, right? So it's not the Fortune 500 equals enterprise, right? Today when we look at enterprise need specifically when we discuss in standing up cloud infrastructure We have small organization Much smaller than Fortune 500 that has the same needs, right? And the main reason is you need to do more with less So cost is a driven also in cloud infrastructure and the same IT Organization that used to provide services now needs to provide as I mentioned IT as a service with the same pretty much budget This actually drives us where are they running so in the larger responsibility of they still need to run and support the legacy traditional Propriety or even investments they they're done in equipment just standing up the old-fashioned databases But they also are and they're also bound to what we call pets We have the notion of cattle and pets in in cloud and and when we talk about pets application They still need to provide services in them And when we see also the adoption moving to open stack or in general from enterprise to cloud Infrastructure we don't see a forklift upgrade, right? It's not that we're gonna replace the whole traditional IT investment we have and boom we graded the open stack and we're done The one of the reason is we still need to provide the services and some of the services are Propriety Software that we built in our IT to support some of the services are still very traditional databases oracles sap etc That needs to provide services and not all the services actually fit to the cloud Infrastructure because they were not cloud-born, right? They have specific AJ requirements for example, etc Why do they do so they support everything to the data center to the end user and service provider for different levels of operate business operations, right and we also have the When we talk there's a notion of frictionless IT all of us are IT users and we have cell phones the mobile phones that we carry that Provide us different Experience user experience when we leave our work, right because we can actually Press a button and get the app we need or wherever it's getting information or on our flight or getting a Social network GPS to get to take us home, etc But where is IT the do the IT already made up to that point to be frictionless ID, but we're not there yet Cloud infrastructure allows us to get there And when we look at OpenStack community support for enterprise So we came a long way in the last 11 releases of OpenStack The expansion of design summits to embrace the high-level of conversations to those of you attended today's sessions And we're just starting now the design sessions themselves There's a lot of key Elements being discussed that are amazing from the table, right every release we're getting better of the good news We have very short cycles every six months We get able to deliver a set of new features and when we look at the enterprise and product manager working groups This is something new that was not there before and our cutting maybe can elaborate more About this but we started to have a working group within OpenStack to actually Tackle the enter the window enterprise need we started to have discussion around okay It's time for us to look ahead at the OpenStack Development cycle and start to see where we're going in terms of driving changes in a much more early fashion to actually Put some product management. Yes, it's an open source product Software, but we still need to put the frameworks One good indication is that just look at the number of attendees In the summit right it's all around six thousand people So if that's just the people attending you saw how many lines of code we have in OpenStack You can do the math right we need that framework to come alive Another side effect of this is the number of drivers who have more devices more hardware That's it's plugged into OpenStack because we have more and more vendor interested to providing and again OpenStack is all about being a pluggable infrastructure However, who's doing the CI? How is the CI being done? The CI is the continuous integration testing For these new devices with OpenStack Hardest certified or not etc. And of course, what does it mean from? the commercial distributions to provide longer support cycle for this a Roll across a train that upgrades and provide updates every six months With that I want to actually zoom into the Dell and Red Hat approach to this so far. So in a sense a lot of our work is Considered to be the the boring things. However, it's the boring thing that matters And I mentioned earlier just only in the last two cycles pretty much We got to a much more stable course. So we got the core right However, making OpenStack repeatable is something that we need in order to be able to do a bit in production, right? I just for me just to stand up today in OpenStack environment. I need to stand up nine different services If I need to do the same thing all over again in my next data center or my next site Will that experience be repeatable if it's not repeatable then it's not ready, right? So we need to come up with a repeatable routine that we can deploy OpenStack that know that works And of course, we're going to do it in the next slide We need to make OpenStack testable. So I mentioned the drivers and the number of complexity arrived with Just supporting all this rich ecosystem and when it comes to best practices and configuration There's more than one way to configure a j. There's more one way to stand up your your deployment Just stick networking for example with Nova Networks and Neutron with all the different configuration, which is the right one And of course fewer snowflakes. So in a sense OpenStack is We we try to get a point where OpenStack just works, right? So what is the delivery model that allows us to provide an open state that does just works without having like a Cardis-Sensei To fix it for us, right? And with that I want to end it off to Arkady To showcase what's we have in the third release of the joint solution We will not try to have a specialized things for each individual Deployment because a it is very hard to make it work You know the first time and then you would like to kind of repeat the process So you share the experience And the second thing we wanted to ensure that you know instead of trying to optimize it for specific individual need We wanted to start with something which is kind of good enough for you know, majority of the use cases And when I say good enough, I don't mean just the configuration of the OpenStack Which is what this community is kind of concentrating on But actually go one step beyond and have you know end-to-end solution which includes hardware The distribution all of the configuration all of the resiliency features and all of the things which we As kind of the end user take for granted, but somebody goes and have to make sure it all works so we kind of come up with kind of base base things as Sean mentioned the June release was kind of the first release we felt comfortable enough to make To make kind of ready for enterprise There's actually the first release which we felt comfortably to move away from novel networking To the neutron and as a first release we jointly created which was neutron based By no means we means that we will support all the different options which are possible even for the networking We actually decided explicitly that we're gonna support only one specific option And there isn't one because that one we tested well enough We know it will perform under the stress it will perform properly under the failure failure conditions And it will deliver the scalability which the customer then expects to do Right from the start you know we call it reference architecture But in reality it is the reference. It's a kind of minimal production Solution we wanted to deliver to the customer and because of that Some of the features like ha is built in right from the beginning You know it doesn't matter how small it is you will have the full ha capability not only ha But it will be active active ha across all of the services in the in the solution We made a very conscious decision about what kind of hardware we would like to have in the solution And we chose the one which is a balance performance wise It's not the you know the the best possible we can do But at the same time we were kind of consciously made a decision of Price performance ratio. What is the right choice for the starter platform? So I'm not gonna walk through all of the different possible configuration but there is one other thing which I do want to bring into into Focus and that is we decided fairly early on That while we will support multiple different back-end storages We actually as a base solution would like to have self as a core cornerstone of the solution And the reason is because that allows us to have a single type of storage Even though we can configure multiple different pull for different purposes But we'll have a single storage cluster which support all of the needs of the solution It supports the sender. It's support the images in the glance. It's support if a merrill storage It allows us to do the life migration. It supports various different things, but at the same time you are It's a software defined storage So it can scale out as the need as you go forward and try to get a larger production solution without any forklift changes or anything else and also it allows us to to To To run it on the base hardware In commodity hardware, which you know you can You can buy and we configure it actually very similarly all the way across the entire platform End-to-end and I'll get to that in a second So one other thing, you know while we can talk about you know here all of the components Which make the solution and make sure everything works At the end it is actually the processes we should put in place to make sure that we have a repeatable process both for the quality of the solution going forward as well as ability to Provide the tools to the field either the customer or the service organization To ensure that what was done and delivered to the customer is actually working all the way across So it's not hoisted on the customer to go and figure out why certain services all of a sudden becoming flaky or What was other problems might start happening? earlier on David was given a talk about One of the work you were doing upstream is we are using we are moving the tempest To be not only the gate in the open stack from which it was designed from date one But using that as a tool which we can deliver to the customer so we can have that as a validator in the field and Has multiple different benefits because customer can you know run it at any time to where it did that everything is alright It doesn't you know we're making sure that there's no side effect after running that and you can kind of run it repeatedly to see if all of a sudden your Your infrastructure which was running fine all of a sudden start getting flaky Okay On top of kind of delivering The tools to the customer to the field to be able to validate the solution in the field Of course we go through a fairly exhaustive Testing process we set up our own infrastructure actually joined such that we continuously testing Every change which we are making on the software side as well as any change potentially we are wanted to verify for the Reference that for architectural change on the on the stamp continuously so we have the you know joined Infrastructure going on all the time and not only testing the simple thing like API which tempest does But we are also testing the full HA capability. We have the full test injection both on the node basis on the on the Storage sides on the switches on the nicks and so on so it's kind of very exhaustive level of testing to ensure that what delivered to the customer is Working notice it. We don't do it for the each platform. We deliver to the customer We do it for the entire solution and you know, we repeat the same solution again and again It's part of the building of solution which have those feature rise from the start We of course provide the guidance. I mean well, this is I mean by definitions reference architecture So this is kind of the base you want to start from but then we can go from there to various other ways of how to improve things either you want to scale the compute you want to scale the Storage you want to scale your networking whatever whatever the needs you have for your specific application domain We have a guidance of how to proceed For various Reason for for the robustness reason as I mentioned I'll get to that on that work in the second. We actually have fairly You know do a redundant network all the way through the system Well, you know, I have seen too many other reference architectures and all of them kind of Suffer with the same problem which were part of the kind of built in the testing which was part of the CI of the tempest by the community and that is every Service becomes a separate network with a separate nick of the view of the view on the on the VM to test it It's definitely easier to test but that's not what you're going to deliver to the customer and Every time, you know, you said oh, okay. Well, I need to connect it to something else Well, you're not going to add another nick to your system every time you want to make change It's it's not going to scale it will cause problems to the kazoo and every time you have to do any change You have to go and update a physical architecture. No way So we kind of went from the beginning saying no, we're not going to do that We're going to have several physical networks on top of those physical networks We're going to have the wheel and specific for each of the flows So you want to connect it to some other device? No problem. We create a wheel and it's Fully automated you can do, you know, you can create your own tools if you want to or you can, you know we'll provide that as a service and You know, you can extend not the physical network, but the flows which you want to have in the network So I'll skip to that We have all sorts of tools to help you out right now how to configure the hardware and actually I'll go one step forward that Actually, we are actually very actively participating with ironic project to make sure that that becomes You know hardware is a as a service so we can configure it It's actually the work is way ahead of what is being done on the ironic We are you know, as I mentioned in a few slides We actually have all of the configuration needed for the solution already built in and what is called RDO manager Which is a joint tool we are developing which allows us to Not only do what ironic can do right now in the killer We should just release but allows us to set up the bias set up the rates set up everything Which you need for every of the node type for each of the function they need You'll do one thing for the storage node because you want to set it up You know for the self in the very one way and you want to set it up for the compute No, but for virtualization completely different way Not only it's different type of the hardware But even if it's a similar type of hardware you will configure it differently for different purposes And that's fully kind of automated work what we are trying to deliver in the next set of the In a next set of the solution coming coming after Let me spend just a couple of minutes and kind of Outline the point what I was trying to make before So if you look at the nodes we kind of classify them in different categories But all the nodes have somewhat similar structure with respect to the networking we have to do do dual port snicks and We bond those nicks to provide you full redundancy both with respect to the failure, but also for the for the full performance connectivity Conceptual network can be separated into the three categories one, you know public ones and which we only the public traffic will be run the private one which is basically for internal communication which will be running and then the The management one which is one gig because you don't really need a very you know high throughput for that And we are price conscious to delivery the solution, which is the best for the dollar As I mentioned before if you look at that you have one pair of nicks and you have five different network sitting on top of that and This is how we Take the wheel ends and making sure all of them are operational on top of that And if you want to have one more flow connected to something else maybe for different type of storage Which you have already in your data center, or you will connect to one of your own internal, you know databases We can provide that without changing the architecture at all the The physical infrastructure, I mean which is kind of not here, but all of the nicks are connected the same pair of switches which are which are ensuring the full redundancy and We choose Right from the start in the configuration the type of the bonding we wanted to do for the nicks to ensure You know the most resiliency and the best performance at the same time Knowing fully well that whatever we'll provide to you is supported you can go and try several of the others Or you can trust us that this is the best one because we tried several of the others and we know where the gachis are So at that I will pass the button back to Sean All right, so When we look at the high availability that our cutting just dusts upon right and you mentioned bonding So part of the validation process we done we actually tested open-stack from top to bottom And guess what the level of bugs we found were just not just in open-stack We found bugs in pacemaker. We found bugs in bonding right, which is two layers below. So Just to come up with the perspective Solution that actually works you need to go all across the testing to make you find the right solution So we when we call this a reference architecture. It's not just a reference actor This is a reference architecture that actually works and it's good enough for production. So there's a big if you go to the upstream and Go and download open-stack and start deploying it as I said earlier There's tons of different configuration, but if you want to stand up open-stack for production This Combination is really really never down. So this is why as Orcati said we found some areas you only have one choice probably that actually works And we talked about pacemaker AJ proxy and horizon virtual IP This is just one slide to Summarize how this actually works under the covers With the pacemaker cloned AJ proxy service where we actually have horizon for virtual IP as well that actually can provide you this full redundancy If we look at the progress we mentioned the progress being done in june and one example is Detect the lay L3 networking right which is the tenant network and where we were Until june release so in june release one big Accomplishment was done is around the VRRP the virtual router redundancy protocol until then we had a single point of failure So if you're the host that's running the network agent goes down every connected Tenets using this service will be pretty much down right. There's no one to resolve The network for them so that actually now Resolved in an old relos p6 and in Juno and now we have an internal agent network Which created pertinent with keep a light process that it's fun and this is out look so we have a Network node one with virtual master virtual router backups as well as Network no 2 and free with the backups with fully redundancy using the VRP protocol but Talking about as I mentioned open stack top down We it's not ending with testing open stack But we also and I mentioned some of the bugs we found and and and the bugs as I mentioned can be found in different layers And open stack is just like an onion right does have a lot of layers and open sec as you know Does not do everything on its own So if I go ahead and create a volume in open stack Opens actually go in colds lipvert to do the dirty work and create so we have a lot of layers Underneath open stack, which is the cloud infrastructure that goes to the hypervisor go to the OS hope that's on the host all the way down to the hardware and We're dealt with bugs in the last year that actually were found in the cloud infrastructure Was open for open stack and when we drill down through the stack it went all the way to the kernel and And when I talk about the coin geneared and integrated from a reloy speed perspective One of the biggest added value We have as read at is actually the fact that we are building open stack on top of rel And this actually allows us to not just fix bugs in kernels that we're not now doing For a lot of years just from the operating system, but we know the stack all the way up We know qmo. We know liberate We know all the pieces that allows us to support the stack all the way down and believe it or not another big thing It brings already to the table is the supported hardware because we're doing to integration with the hardware drivers Already from the operating system let alone. We're doing is as well for certification for certified plugins with open stack So if I'm a customer and I just like it how many of you drive cars here? Just ready to hand all right. How many of the cars you you're driving use computers today? All of them right what happens if your Brake system in the computer car computer goes down and indicates a Problem you're on your way from work from work with home, right? Where will you go? Would you stop at any shop on the street? Please fix my computer car because that's a breaks in or you will go straight to the manufacturer to fix the computer This is pretty much similar right you have to know if I have a kernel problem in my cloud operation Who would you go to? So this is why we think we're bringing a big value add Integrating all of this and it's not just being able to support the code It's actually be able to impact the code today We're doing performance enhancements for example at the KVM level at a rail level to impact open stack So we are able to impact the whole stack To get it better. So I think that's the message of the co-engineered and if you connect that to the Co-integrated with the hardware as we done earlier and Arkady was great to present Then you have a good story and if you tie it all together to what we're trying to achieve This is how we we actually be able to deliver something we can actually Not just stand out that works, but also provide the whole support that is required and just to to Some of that is that we offer today free your support to open stack and as you know open stack has a six months support cycle a release cycle and Typically it forces customers or not just forces, but what do I do? Do I upgrade to liberty or not? Can I stay in Juneau or? Diablo whatever version I jump the train of open stack app So it surfaced a lot of needs in terms of all right. How do I? Get my patches if I stay on on that release We'll be able to get similar features backport, etc. And of course Upgrades which is a huge pain point in open stack though. I'm going to touch upon so I'm just going to talk about some What what's expected right what we have next so we touch upon that in Juneau with the L3 VRP support Nova one of the things we're focusing. This is specifically in relo is p Based on the enterprise needed we talked about the legacy applications that needs age a which is not there specifically today In open stack for them. So one of the things we're practicing now is actually to leverage pacemaker as well for Enabling instances high availability. So if I'm a user, I need to be able to have sorry Automatically I need to be able to detect the failed hypervisor and we great all the running instances To another hypervisor automatically and we're not there yet from that instances perspective. This is something we're invested looking ahead To what's next so I talked about Free topics that we want to focus on that we intend to find so I talked about high availability For services. The other two is rolling upgrades and deployment as I mentioned upgrades is a big Thing in open stack today, right? How many of you have upgraded your own per second environment? Good luck So it pretty much involves today stemming up another cloud environment and migrating your your workloads We are looking to actually provide a much more robust way to do it with minimal downtime And I'll need to do in place upgrades as well And the way to do it of course is using continuous integration and new methods that are not there In terms of deployment, we're looking for the easy button installer that enterprise needs, right? So how do I simplify this whole process of deployment with open stack in a way? So I'm gonna skip this and just talk about the development Life cycle so the area we identify we want to focus on that when we say development life cycle is this deploying open stack That's phase one once they deploy to have different set of problems. How do I manage open stack? What how do I know remember the car analogy I gave earlier? So I have a Problem, how do I know I have a problem in my cloud? Where is the alert in the dashboard? And if it's there, what do I do to control in the bucket? I need to provide a PA may open API To make other management tools. I need to be able to add or remove Compute node of storage node to my environment and last thing how do I introduce updates to my production environment, right? I'm not talking about just upgrading from Gino to Liberty or whatever but patch management for example, right? How do I do regular updates? Take security bucks right zero-day security bucks that you need to deploy in your cloud environment So deployment is more than just the initial open stack installation It's also deals with the deployment cycle as I mentioned and one of the things we're going to introduce this summer With relos p7 is well a speed director Relos p director is a new deployment and management tool that is based on triple O That allows us and also inspired by spinal stack Project that allows us actually to deal with not just the deployment, but also the management aspect So if you're interested, this is already available in the rdo Community edition under the project named rdo manager So if you just go to the rdo website and look for rdo manager, you'll be able to already start look at it There's demos. There's how to install how to contribute It's available already today for you to start work of it play with it and with that I want to zoom back and Summarize what we've done so far. So so we saw the enterprise needs We saw where open-sec is and you saw also what approach we took actually to bring it out together In a manner that we can support you can deploy and you know You have the quiet at night that this will work and I'll break because someone else has tested for you And I think that's some of the things we were able to bring to the table And this is actually available today. You don't have to wait to liberty released in order to deploy open-stack That's a good news, right? So if you scan this barcode, you'll be able to download The Dell and reddit cloud solution reference architecture this reference architecture architecture is not just one as our cutting mentioned that we started You can start with a very small deployment Go to a much larger deployment that the for example of free racks today And and it allows you more than one configuration for example take storage as as our cutting mentioned Etc. So with that I want to quickly open it up for Q&A And if you have any questions feel free to use the microphone up here and we can try to answer One of the requirements for enterprise is that the new open-stack environment has to work with existing infrastructure does your White paper or reference architecture has section that describes how you how you set up a new open-stack architecture that works with All right, so there's two answers level of answers one of them as you saw we are very perspective and that what the RFR Uses right the network this the storage we're using etc so in a sense it depends what levels of components you would like to change and But if you're looking at the network models for example, they're pretty much straightforward that would you don't have to change? So if you're looking for specific equipment or specific storage beckon It all depends on the question and I would look if you need to change anything with there mainly look at the if it's certified or not and From a driver perspective, so because we actually tested everything that showcases here is well tested so we bring you that The quiet so if you change any parameter that you need to be able that whatever you're changing is certified and tested So we have open-stack So I'll just come add one more thing to that So when we in you know when we talked about the OSP director and one of the capabilities bringing up When we talk about upgrade, we are not talking about upgrade just of the open stack. We are talking about great of the operating system We are talking about various components be it self-component be it pacemaker be it various else But going with the support of the ironic as a part of that we're also upgrading the bias Of the underlying hardware to go along with you know closing the gaps and making sure we have the upgrade of the solution end to end Not the piecemeal. Okay. Yeah, I provide you upgrade with the open stack. Well, okay What about the rest of the solution? Well, you know, we left it as a size to the user or you know You go to your vendors and let them upgrade it and pray that when they upgrade that part. They don't break the other one All right, hope that answers your question. Any other questions? All right, if not, I want to thank you all. Thank you very much for staying until the end. Yeah