 Thanks everybody for coming We'll go ahead and get started if that's okay So how many of you guys are really enjoying OpenStack? It's a trick question how many of you guys are asleep So my name is Michael a day. I'm from Heela Packard. We also have Eric sacks from Oracle and Glenn Foster from Oracle as well and We have Nigel Cook from Intel and today we're going to be talking about this interesting very polarized debate about pets and cattle and how we can how we can add some current best thinking to it as well as make OpenStack very safe for these kinds of adorable little animals that It's ironic. So this is Eric's cat and And whenever I saw this picture I said did the cat really look at you like that? Because I mean Correct me if I'm wrong, but that cat does not have the typical disdain that most cats have for their masters Correct. Doesn't this look more like admiration than disdain? So he must be doing something right with his cat and I suspect that's translating into Stuff that he's doing right in OpenStack, although the correlation hasn't been established yet. So How many people are familiar with the the argument around pets versus cattle? Okay, if you're you're in the wrong place you need to leave no, I'm just kidding We're not that polar here So pets versus cattle is a great argument for Building a community because it's it's very elementary. It's very binary It helps people understand the concepts But it's really bad if you it's really great for illustration But it's really bad for a community and the reason for that is because this is a very polarized argument And it and it really doesn't have to be Pets versus cattle is illustrative of the underlying Opportunities that we have within OpenStack and To approach this argument in the polarized way and that is currently being approached It's not a bad thing, but it does it does create an interest gap and it creates a an Exclusion and in reality OpenStack will be most successful when enterprises deploy OpenStack And I mean deploy OpenStack for more than just easy web workloads When they deploy OpenStack for critical workloads when they start using it for their their infrastructure their the spine of their organization and It is very well suited to do that in many cases and there are clearly some opportunities that remain but You know we want to begin depolarizing this argument as much as possible and understanding what the underlying implications are of this argument and The way that I generally think about this now is that it's no longer about pets or cattle It's more about whether you have a highly managed workload Or you have a workload that requires very little management a lightly managed workload so heavily managed workloads I equate to pets and Very lightly managed workloads. I equate to cattle and I think that if we if we can understand those those ends of the argument and the stratification within that spectrum then we can begin to address the shortcomings Inside of OpenStack that make it less acceptable for certain types of workloads So again OpenStack will win as infrastructure when it begins being deployed in areas of increasing criticality and In order to do that Will pick up workloads that are increasingly important to the enterprise these workloads Have some general properties they They become increasingly highly managed They tend to want to scale up instead of scale out They don't support the the architecture that's associated with cattle based applications and Or cattle cattle designed applications They have performance and reliability tied to the underlying infrastructure more closely And they might not be as tolerant of the mechanisms for for dealing with single points of failure as Cattle are and so it's this is not that they are bad They are they were designed with a generation of architecture that existed prior to what the current gen is People have been developing applications You know in various different schools over the last 25 30 40 50 years and these applications Have the architectural schools of developing these applications have moved on and so as as we start looking at these Some of these are applications that have been continuously maintained inside of the enterprise They still offer a supreme level of criticality for the enterprise and so there there's this principle of do no harm The art and you don't want a significant rearchitecture of an application if it's servicing a critical workload And and these applications don't just exist within the enterprise they also exist within the carrier space Within the service provider space So every or every audience has its own set of pets that is trying to maintain So more heavily managed workloads Happening in a variety of dimensions. We're going to talk about these dimensions throughout the presentation But I would posit that there are storage-centric workloads So we have Oracle Partnering with us on this presentation and obviously it's it's a pet breeder, right? No, no, no, not pejorative like that in a pejorative But then you also have network-centric workloads. So how many of you guys are participating in NFV working groups or working around NFV topics So obviously the network can be the source of pets as well Storage can be the source of pets network can be the source of pets I'm not sure about compute, but I want to fill that in so if anybody has any great examples of how compute is the breeder of Pets and that would be great So let's talk about pets and cattle. So So pets are born Because you need to service a workload you need to buy some hardware and You're going to buy that hardware in a very specialized way around that workload and You're going to rack and stack that hardware and that be and that is the beginning. That's the birth of the pet Right You install the servers you configure them you deploy and test all of these infrastructures are supremely specialized around The pet so the pet is not just the application The pet is the infrastructure to host the application The pet is all of the surrounding accoutrements that go I'm practicing my French they go along with That that application that make that application able to service the need that the application was designed for And then you have the ongoing goo, let's call it technical term Around the pet you have to buy bedding for the pet you have to feed the pet you have to walk the pet you have to pet the pet You have care and feeding that goes along with the pet You have to you have to manage the pet you have to monitor the pet you have to examine faults in the pet You have to understand when those faults have occurred respond to those faults very quickly Because this is a pet and you don't want the pet to die I prefer to call it again a highly managed workload. So this is highly managed by definition So failure in any component of the pet and I would argue that some pets are tolerant or some Highly managed workloads are more tolerant of failure in some components than others But this again is a specialized problem. So you have to understand the nature of the pet and you have to respond to the failures within the pet so anatomy of the birth of a cow or anatomy of a cow So or cattle depending on your perspective or or lightly managed workload, which I kind of prefer So you start with a workload you don't start with hardware You don't really care about hardware. You don't really care about infrastructure In fact, you don't really care about much of anything other than the application itself So this begins with a workload. So the workload is the birth of a cat of cattle and it's deployed across a whole slew of virtual infrastructure and It's deployed with very little dependence on underlying hardware and it's deployed with very little dependence on state So the cattle exist Without any dependency on the underlying infrastructure the cat and this is a this is a very well architected cow Okay, so bear with me on this They they have no dependency on state. They have no dependency on storage. They have no dependency on network They're just there if they fail You replace it you replace it with another instance and you resume Ideally, these are scale out. So you're not You're not dependent as you're not dependent on virtual infrastructure scaling up You're dependent on an array of infrastructure supporting the workload So if you need to add if you need to add more scale You just add another VM you add more virtual infrastructure and you continue on and Since the cow since the ailing cow the sick cow is Effectively stateless you don't lose anything Whenever the cow dies You have very little investment in the cow. So So cattle so lightly managed workloads everybody how many of you guys are transitioning enterprise workloads to OpenStack today Two people Am I in the right session? Is this the right? Oh, okay. Just checking. So So Two of you are transitioning enterprise workloads to OpenStack. That's awesome. You're probably dealing with You know, you're probably doing a lot of introspection right now You're probably looking deep into your soul and thinking is this really What I want to do with my life Probably thinking where did I go wrong? I have this thing. It's not the thing that I want I need to change the thing that I have to be the thing that I want Right. Am I pretty accurate? We've all been there. Okay. It's a dark time. We're gonna see you through it because guess what happens The thing that you have is perfectly acceptable We accept you We accept your thing The there's nothing wrong with having a pet In fact, if you didn't have a pet, then you probably wouldn't have a business So we we want to support that and there are a variety of ways that we can So and you're gonna have these things for a long time because in the future You know, most people envision a future of nothing but cattle But pets are essential Highly managed workloads will always always exist. There will always be scarcity There will always be management required for these workloads You will always have to have some degree of infrastructure capability specialization because that's where performance comes from That's where capability comes from if you could deploy on any generation of processor any vendor of processor any vendor of network any Capability of network then that means that you're targeting say it with me lowest common denominator and Guess what comes from the lowest common denominator? Poor performance So you will always have various levels of management Associated with these workloads and that's not a bad thing The bad thing is whenever you have a dependency on something that doesn't exist in the system And that's the reason that we have Nigel talking about when the enterprise and the work that's being done there And that we have the rest of the presentation, which is far more important. So I'm going to continue So meanwhile back at the enterprise Clouds are forming right people are deploying clouds because clouds cool Clouds a new resourcing model clouds a new management model cloud represents a tremendous advantage to the enterprise Cloud represents the ability to fulfill business drivers for compute in new ways and address concerns that exist within servicing those types of business drivers, and so many enterprises have and will deploy Open stack and this is important Because it allows them to fulfill their business objectives and as they continue to grow They get all the benefits that open stack brings Elasticity scale all the benefits of the hard work that's being done within the community and they decouple their applications from their supporting infrastructure You want to take this? Yeah, sure. All right. I'm gonna hand it off to Eric because you're probably bored with me anyway No, no, thanks Mike No, I think I'm okay. Okay cool This is gonna roll off. I sense it there you go. Okay, so to recap I think I think the position that we're seeing as we talk with a lot of enterprise customers sort of They're sort of caught between the promise and the interest of moving towards cloud and all the all of the benefits They see with cloud all of the agility benefits the ability to scale up And to have more of a centralized self-service infrastructure But at the same time they're sort of asking these questions, you know I've got you know the vast majority or some key subset of the workloads that I run in my enterprise Are pets and so what am I supposed to do? What's my migration path forward right and you know per the slide that Mike just talked about there's sort of the question of well Yeah, so should I be looking at taking my pet application and you know try to try to sprinkle some kitty dust on it And somehow form it into into cattle over time or look at look at rearchitecting it Is that something that's possible and you know in some cases it might be But you know this is this is kind of similar to What I think you know it's similar to sort of trend of you know where years ago the vast majority of applications We're all single-threaded and with the rise of multi-core and multi-threading You know you want to sprinkle on the multi-threaded dust and all of a sudden get some concurrency out of your application And that's hard to do in many cases you have to sort of rearchitect the application in a very fundamental way In order to do that and you know as Mike was mentioning there's a lot more to Pets and cattle than just the workload itself There's the infrastructure that's also sort of been designed or baked in when when considering and conceiving the pet So it's difficult. So I think that you know, there's one of two things here either You know as this plays out in the future You know either it will happen over time and folks will be successful with taking you know the applications that are implemented with pets and Being able to re-implement and re-architect those and as cattle and I can I can believe for a some Some set of workloads that will actually happen or maybe for some it won't ever happen, but in either case It's gonna take time So, you know, this is really kind of the crux of this presentation Which is you know really sort of putting forth the notion that it's not just pets or cattle It's both And both pets and catables sort of need to be brought into scope here for for open stack so years ago the matrix I Reminded of this scene the wisdom of the the young boy as he's talking to neo and for whatever reason I got to be thinking about this and You know, it's sort of you know answering this question around the conventional wisdom Okay, well if I can't re-architect my pet can I go ahead and move my pet into the cloud? And so to the the conventional of the wisdom around this today seems to go Something like this, you know, do not try to I won't do the the little accent But I'll at least try to say this as best I can do not try to run your pet reliably on cloud Infrastructure as a service that's impossible instead Only try to realize the truth. There is no reliable infrastructure as a service Then you'll see that it's not the cloud that needs to be reliable. It's only your pet And this this seems to be The belief that a lot of folks abide by now they just sort of okay. Yeah cloud is this way, right? It's a sort of this Unreliable infrastructure or it's this rely it's this infrastructure That exists and it's it's great for cattle and it's perfect for scaling out, but it's not really a suitable home for For my pet, but I think the thing that we would say Is What if we told you we can build a better cloud infrastructure so that pets can be managed to and Mike makes the point that you know what we're in a lot of cases what we're talking about our our SLAs some Applications have more SLA needs Than others, you know pets having more and cattle having less So this was this was also spoken to at the Atlanta summit Toby Ford from AT&T got up on stage And you know this was acknowledged in his talk You know and very clearly, you know, Toby was mentioning about how They've been one of the the leading enterprise customers that's been making a lot of progress towards Moving towards cloud adoption in the enterprise But they sort of acknowledge You know the one thing that he was really worried about when talking about open stack and the open stack ecosystem is That the scope is too narrow. It's not just really about cattle applications You know Toby says the pets have to be handled and they have to be managed and open stack really needs to bring pets Into scope more clearly So I'm gonna hand it over to to Nigel This really is a tag team episode here So I want to talk a little bit about when the enterprise how many people have heard about this this initiative within the Within the foundation some people have so let me give you a little bit of history on this part of what the open stack board does is Survey users of open stack and so Tim Bello soon publishes a six-monthly Survey and gets back comments about what the structure of folks clouds are What they're doing in the clouds and and more importantly comments about their clouds and about Open stack and the work that the community does and and based upon that feedback the the open stack board met and realized that they needed to kick off a initiative to look at the gaps that were Basically limiting enterprises from adopting but the community work and so this was kicked off in the Atlanta summit and Since then there has been a lot of work going on with a lot of people from a lot of companies looking at the feedback looking at sort of maybe anecdotes from their use of the cloud within their organizational or for their customers and and and sort of getting to a curated list of Here are some top issues that we really need to fix if we want to Accelerate the adoption of Open stack into the enterprise The the work groups have got sort of split off into a whole lot of application areas and as a result of those Those working sessions. There's a series of blueprints that have been proposed for kilo They're being discussed in the design summits over the next few days as well as some things sort of crept into Juneau as well There are sort of demos at various booths of some of the blueprint proposals which are aimed aimed at making Open stack more amenable to enterprise adoption and in those areas. There's a lot of items which sort of cover this this pets pets initiative The demos by the way, I'll give a slight plug here I am from Intel if you come to the Intel booth a number of the items here You'll see you can see demonstrations of those blueprints and some of the things that are being proposed at the summit here is a list and It's not so much of an eye chart but When you get the slides you can search for all these blueprints if you're interested in Both the topic areas and the proposals from various teams and how to address the problems that have been highlighted and you know if you look at Some of these areas you can also see things that were Mentioned in the keynotes either yesterday or today So I remember if you remember the time Warner presentation and the BMW presentation about Live migration of a VM with a volume components about live upgrade And and you know these and VM availability these components are areas that really Scored highly in some of the surveys that were sent out and people are looking specifically at an effort within the community to sort of solve these problems to get to a you know a new More stable more enterprise friendly Infrastructure and one that then is suitable for running both pets and cattle So, you know, I'm gonna sort of wrap up here We have time for a little bit more in terms of questions from folks So I want to just really recap about this You know if you if you want to build a solid house You build upon a rock and so the infrastructure and the infrastructure that you design for your cloud needs to be Needs to be a good one and in fact if you look at The the large public cloud providers and their infrastructures. They're actually very well designed actually use high-end components because what they're trying to do is get extreme density extreme performance And in fact, it's those same architectures which are also able to run those scale-up applications the carriers are Have quite demanding service level requirements as they look at the infrastructure They need to support they came back with these same sorts of things that actually really the enterprises have been saying as well You know, how do you upgrade? You know, what am I going to do? I have a firmware patch. How do I drain a machine to do that patch? Without disrupting everything in the system the same requirements coming back again and again From a compute perspective, there's a variety of different hypervisor models that are available KVM is great So is VMware. So is hyper V. So is Solaris They are all perfectly valid hypervisor choices that we need to support and and Which are being supported down the community to put to keep Enterprise workloads running happily and transitioning easily and then high availability is really about In my mind having an infrastructure which is a minimal to To common maintenance operations maintenance is the biggest cause of failure of an application And so if you can't maintain your infrastructure in a way without killing your cloud for six hours Then you know, you haven't got a cloud. That's really suitable for anything So that's our presentation Maybe Filling blinded by this light here If there's some questions of the people have to ask now we can jointly try to field them You guys are too kind. Oh Come on, there's got to be at least one question on the back Go ahead. So the question is for The assumptions stated in the slides was that cattle are largely stateless and How does it relate to? To applications that are dependent on new types of state Like those delivered through trove and other services like it. So the question further stated is When you have state state based applications applications that are relying on state are they pets or can they be cattle? and what I would argue is that pets and cattle are is a binary concept and That there is no cattle There is no notion of a cow and there is no notion of a pet that they exist in a continuum and That state is one of the factors that causes a workload to be highly managed And that concept has been socialized as a pet and So pets don't just come from state But they come from a lot of different factors. They come from dependence on hardware They come from dependence on other applications. They come from dependence on State they come from dependence on storage and all of these all of these Dependencies in aggregate are the definition of a pet Because the more dependencies that you have the more management you have to maintain the workload and so so I'm I I Agree with your point that state can move around But the notion of state the the requirement to have context Is a is a it creates a dependency which creates a management activity? Go ahead Want the microphone? Can you hear me? Yeah, so what about the idea that each? Applicates Applications that are suited for one type of hardware environment should remain there so an airline system that was built on the mainframe maybe is best run on mainframe and ERP SAP or Oracle is best suited for physical servers and maybe virtualized and Email and other three tier applications could be virtualized on a virtual You know VMware Zen KVM platform, but the new Open stack is designed for status cloud applications like streaming like Netflix like and Each yeah, you you you see here in Paris a small a small Economy car is where suited for the traffic in the parking, but if you want to drive down to the coast You may want something comfortable or a comfortable bus I think I think one of the I think one of the points of the presentation And I think we're seeing this with open stack, and I think this is one of the things that's Really quite appealing about open stack is its flexibility in enabling cloud deployers to Deploy a choice or a different mix of different technologies within the cloud itself And I could see it being a totally viable path where if you have a particular pet that has A particular need around infrastructure Being able to leverage the heterogeneous support that open stack offers for systems and virtualization To be able to take that system and bring that in the fold of the cloud, right? because I think I think that's sort of the the the false choice that so often exists which is you know either Either I can have the benefits of cloud infrastructure Or I can you know have my workload and have it have to sit outside of the cloud and not be managed And I think that you know as The open stack ecosystem enriches and grows and has support for a more diverse set of virtualization and system technologies The chances are going to be increasingly good that for a wider variety of enterprise deployed Applications that open stack will be able to manage those as well. I agree. Yeah, absolutely Thank you for coming guys. We really appreciate it. Yep. Thank you