 All right, there we go. Hello and welcome to NFV for Enterprise. Thank you all for joining us. We're going to be talking today about the role of network functions virtualization in the Enterprise. We've seen a tremendous amount of interest in NFV over time. A lot of this has really started to shift towards what operators and telcos are doing in terms of where a lot of the focus has really shifted, a lot of the effort, a lot of the energy, but yet there are tremendous use cases within Enterprise and one of the things we're going to explore on our panel today is where is all of this going and what makes sense in terms of where we are for enterprise-based NFV. Let me introduce my panelists. We have actually, we're in exact order. This is awesome. Pineda Kandia from Murakura, Jared Goth from Nuage, Nila Jax from the Open Daylight Project and Scott Snidden from Juniper Networks. I'm Eric Hanselman from 451 Research and I wanted to start today by first getting a quick show of hands in terms of what your interest levels are, what your focus is and more importantly what your experience has been. So to begin with, how many of you are focused on enterprise networks and enterprise network functionality generally? How many operators, telcos, service providers, traditional ISP kinds of folks? Okay, so we've actually got an interesting mix really across the board here. So I'd also like to figure out in terms of what you're looking for here because we've got a lot of subject matter that we can cover. We want to ensure that we're helping answer the questions that you've come here for. So how many of you are looking to implement enterprise NFV functionality today? How many of you have virtualized some components of the capabilities that you have today for firewalling, load balancing, VPN capabilities today? All right, so we've got a reasonably good mix in terms of where we are. A little bit of background, how many of you are familiar with the Etsy NFV architecture? Oh good, all right. So I think that says from where we are, we've got enough to be able to move forward with a little bit of background. If we need to dive into it, I've got some architecture slides that we can move to if we want some background. So we start out with that leading question of, where are we in terms of NFV in the enterprise? And maybe a little background in terms of how does that reflect what's happening on the carrier and operator side? And a little bit of background in terms of perspectives. I'd like to get something from each of you in terms of where you are. Since I'm holding the electric talking stick, I'll start. So I spend my time at Juniper Networks. I travel around quite a bit and I talk to service providers and enterprises. And I find myself when I talk to a carrier, we get into these NFV conversations. It's less about SDN and the plumbing and more about how they actually start using this in their environment. And so what AT&T and others have been talking about this week is a good example of implementing some of these NFV enabled services. But then I go talk to an enterprise and if it's an SDN conversation, we're talking about cloud and data center and virtualization and maybe how they move to OpenStack. But then there's always interest and it might be a sales guy trying to sell a product or maybe the customer has a real problem of, well, there are a lot of network administration problems in a large enterprise network with branch or campus or provisioning load balancers in the data center or whatever that might be. And so that got me thinking, what are some things we can learn from this telco NFV movement and start to apply to those use cases. And certainly OpenStack and others have been doing things like firewalls of service and load balancers of service, VPN as a service, which by the way, VPN as a service for an enterprise might mean something different than VPN as a service for a telco, but are there things happening in the industry around standards and the Etsy reference architecture that we might be able to apply and make a bit more easy to use for an enterprise? Or are the organizations in the enterprise just broken and divided so that the campus guys don't know or even care about what's happening in the data center? So that's kind of where my thinking has been lately and I just wanted to sort of explore what's possible and how we can start to align some of these activities. Cool. Well, to add to that, what I would say is when you say enterprise, the first question is who do you mean? Because a few years ago I went to Nebraska and went to the Nebraska Furniture Mart. They're an enterprise. They're a pretty big concern. They have lots of IT. Their needs are fundamentally different from Boston, Reuters, United Airlines, Baidu, and a few others. And so when I look at it, I think that there's almost three categories of companies out there. They're the carriers who are service providers and the network isn't just important to them. The network is their business. On the other extreme, there are people for whom the network is one of 10 things in IT. It's important it's there, but honestly it's not the top concern. For those people, honestly, my experience is NFV and even to some extent SDN is nowhere in those companies unless they have a very specific use case that necessitates it. The interesting part is really in between, because if you look at someone like United Airlines, you might think the network matters because that's how you book your ticket. However, when they misconfigure a switch and the network goes down, planes don't take off. So it really makes you think about how important is the network to a company that you wouldn't think that the network is part of what they do. When you think of someone like Thompson Reuters, again, while they're not a service provider in the way you might normally think about it, again, if the network goes down, they're no longer able to sell what they have. So you've got the network isn't their business, but it is tightly coupled to their business. And this is where I think that we're seeing the most interesting cases of NFV in enterprises. There are some critical differences, but this middle group has the scale by which it makes sense to invest in being able to change the way that you do things, to invest in the DevOps mentality that you need, to invest in open communities which is a little bit harder. And I think as we talk more, when I'm talking about enterprise, most of the time I'm really talking about that middle segment. I would agree. I think that's a very good way of kind of segmenting the discussion because I certainly agree that you have, there's a group in enterprise for which this is very relevant. And I spend most of my time with a good set of enterprise customers that fit into that category. And I would say that from what I see majority of the time, it's how do I virtualize the firewall? How do I virtualize the load balancer? And my two main concerns in doing that are actually probably the biggest concern there is I want to get away from large appliances where configuration management, change windows are incredibly complicated. So if I can break this up and I can simplify that and I can do things on a per business unit basis or a per application basis, my life as a network engineer becomes a lot simpler, right? So what I see so far is the NFV use cases are fairly narrow and furthermore within sort of a more of a narrow set of enterprise customers. So we at Midokura mostly deal with enterprises and public cloud providers. And like Scott said, most of what we see is, and I hesitate to call it network function virtualization but we see a lot of LBAS and firewall as a service and VPN as a service. Now I don't qualify these as NFV because they have their own interfaces, they have very use case specific ways of dealing with them. Yes, the virtual appliances, but basically customers can interact with them directly. They insert them into the virtual topology directly so they don't share a service chaining approach or we certainly haven't taken that approach until now, although there's discussion of doing that. And therefore, the way they're managed is also from an administrative standpoint. They're managed directly by the user. They're not inserted and managed by a different level of administrator. So those use cases, and I'd be really curious to hear what the audience thinks about those as a service functions if they count for as NFV. What we're seeing more recently is in the security domain, there's a case where there really is network function virtualization. So Minokura has a partnership with Intel. Intel has an open security controller which is an orchestrator, so an NFV orchestrator, but specialized for security. And in that space, so it's doing all of the things that the Etsy architecture requires, but it's orchestrating appliances and it's not a generic orchestrator. And what I find interesting is we're just getting started in terms of where do you place the appliances? At what granularity can you launch the appliances? What happens if you have a lot of different policies or a lot of different appliances you want to launch? The questions that come up are things like where do you place? How many instances do you place? Is the latency tolerable if I have to go off box to reach the network function virtualization? So I think from our viewpoint, we see a lot of the question of NFV for enterprise being around security. Key points. It sounds like what I'm taking away from what each of your positions are is that enterprise really winds up being this subset of a lot of the things that operators are considering in a lot of these projects that are looking at a much greater scale. So yes or no from each of you is that is this just a subset or is this an automation problem? What I heard each of you saying is that enterprises are only tackling small pieces of this and probably not at the scale. I'll jump right in and say probably a smaller set of requirements maybe because the use cases are more narrow but not entirely a subset. You're calling out that firewalling and load balancing is not really the NFV because it's typically existed in its own appliance. Is that something where enterprise isn't ready? One thing we'll see I think is that all the techniques and the technologies that we see being pushed by the telcos, those will help. So pinning to cores, all the modifications we need to do to the appliances and to the SDNs to make packets flow faster. That's shared across the two use cases. One thing I'll sort of throw out there that I think will help both communities and is not being done very much is true virtualization of the appliances is not being done. So the appliances are too big and therefore you have to be very careful about where you deploy them, how many do you deploy, you might have to go off box if you can't afford the cores. And to your point, management is going to be done independently in the rest of the network infrastructure. It's not going to be deployed as part of that. I don't know, Scott, I assume, nodding vigorously is that... Yeah, that's exactly where I was going with this and thinking about it. I mean, right now what we're seeing when we virtualize a network function is that somebody took that big firewall and then feature for feature function for function tried to squeeze it into a virtual machine. And it sounds like from our audience that's probably what a lot of you are doing out there. You have a virtual edition, you've got those pieces that... That works well that has its application. What we're seeing a lot of conversations happening in the Toco space is yeah, there's a lot of development around pinning the cores and using SRIOV and DPDK to make that thing faster and faster, but then it's sort of starting to look like that big box again because instead of a, you know, branded box from one of us, now I have a Intel box. But that's all it's going to do is that function because I'm locking that thing down. I think what we're starting to see a lot of interest in is more of a cloud-native architecture for these network functions. And instead of having one thing with one core moving packets at 10 or 20 gig, we're going to have a distributed architecture that's load-sharing across many, many cores that might spread out across the data center. And then that becomes an orchestration problem, a resource management problem that looks like cloud-native as the NFV ecosystem evolves to that as the management tools move away from managing my load balancer to managing my distributed application. There's probably a lot that we can apply to these enterprise use cases, just from that development. Well, so is this a matter of the enterprise application architectures really being built for an expectation of there will be boxes or, you know, virtualized concepts of boxes as opposed to being able to break apart a lot of the functionality and distribute it along with the application? I mean, that's, I think, that's a big problem on that bigger aspect. Yeah, and again, I think you have to, there is no one enterprise. There's a big range of them. And I think one of the differences I've seen in enterprises in general versus service providers is service providers are quickly realizing that having a siloed infrastructure doesn't make sense, right? People trying, in the early days, you build OpenStack on ML2, right? And that was your cloud. And you had your network, and those two teams were completely different teams, with different books for how to manage them. And, you know, if you've been walking around any of these sessions, what you find is that the AT&T's of the world, but even the much smaller ones, are realizing that they need to bring this together. And that SDN and NFV aren't two separate things that happen to share one letter that they really need to work together. And it is about, all of this is about making our network more programmable. And having a more agile, more programmable network is really important in managing your network backbone. So, you're doing cloud, and therefore these are all tools and techniques that you're doing together, and the question is how do you lower complexity? How do you get out of siloed approaches where different teams are doing different things and try and have one way? I think that we're seeing the same thing with a larger, more sophisticated enterprises. It's probably a year or two behind. You asked this question about is it a subset? In some sense, yes. I mean, it's clear that something like it's a subset of different enterprises. So, there are whole parts of NFV you can slice off and say... No IMS gateways, no... That doesn't... On the other hand, there's things, I was just recently at Baidu just outside of Beijing. And if you haven't had a chance to look into what these second generation Chinese internet companies are doing, it'll blow your mind. All white box, 90,000 ODM across 30 data centers. And they're building a second generation network. They're taking what they learned from what Google did and they're building their own using open source components, getting support and help. And so I'm seeing... And it's not just them. A lot of these enterprises for whom the network is strategic are doing really, really interesting things. Tencent is doing BGP route optimization all the way down to their best customers. These are the things that you've never seen. The service provider actually owns the route to the customer, so it's not relevant for them. So I actually think what we're talking about is a Venn diagram. There is a set of things that are common for the most advanced enterprises and for service providers. And there are some things that advanced enterprises do that service providers don't do, and there are things that service providers will do that advanced enterprises won't do. I guess I would maybe call a trickle down NFV. There's a lot of stuff happening at the telecom layer, that enterprises will be able to cherry pick as it becomes mature and as the use cases present or present themselves. So I think it's all good. I mean, I think it's all great. I think the majority of enterprises, even the sophisticated ones, this isn't the only thing they're working on, right? They've got a ton of stuff to do and to get involved heavily at this time, I think it's probably more than they're willing to invest. So how does that trickle down actually take place? What's that consumption model? Is that something where the platforms that they're working with? Is that something where you start looking at the virtualization platform, a networking layer which two of you have certainly got a vested interest in? I guess I would say I'm not sure yet. I think it's potentially a little too early to tell what that really looks like. I do think that there are elements of this, maybe security and others that will get picked up by folks like Minakura that are providing the networking fabric, the plumbing. I think there are functions that are very logical to be brought into that type of technology. But I don't know that I have an answer for it yet. I think it's still pretty early in terms of what's happening at the operator layer and I think there's also potential for operators themselves to provide that functionality as a service to the end customer. They'll take the complexity and present the service. Maybe I'll add a little bit to that. My view is that yes, some of the technology will trickle down from the telco to the enterprise, but I think the enterprise will really take a different approach. So the generalized orchestrator for your network functions I don't see as much value in the enterprise. I think the enterprise by and large wants very clear models to do things like load balancing and firewalling and layer 7 firewalling and IPS, IDS. And I think we've already built these. So what I mean is the enterprise wants an open stack project. Now is that project and that project should be very specific to its use case? Now is that project something on top of the network function virtualization that we have today that's generic? But I think that specialization layer is very important. The other side is the in my mind what does network function virtualization mean? What are the problems that are different in network function virtualization versus network virtualization? We've been virtualizing BGP for a while. It's a control plane function. I don't really think of it as network function virtualization because we don't chain it. We don't have the performance problems but you might think after all it is a virtualized router or a virtualized router control plane. But I think the really hard problems in NFV are the performance problems, the placement and life cycle management problems and then that layer of glue not just to deploy but also how do you configure? And so configuring the load balancer or the software load balancing instances inside the load balancer coordinating with the clustering technology in a distributed VPN for example. All of these are very specific and will require very specific APIs for the callbacks so you configure your policy in your network function and then you configure the service chaining and the service chaining might use load balancing and so the interactions will be very different if you're using load balancing or if you're not. Well, but that starts to get back to that narrowed use case of the operator model because if you have to do chaining, if you've got to stitch all of these things together as part of the orchestration, you're now starting to do all of the pieces that the operator has to do, right? Right. And actually I wanted to get another quick show of hands. You called out one of the projects for management of the actual NFV instances themselves. How many of you are using Tacker currently? Well actually, let me start out. How many of you are aware of Tacker? Your term? Oh good, alright. But even with Tacker, what you've got is the environment that's going to help you manage what that instance is. You can hopefully keep it up and running. You can do configuration management. You can do some definition around it. But it's not going to do that stitching of the data flows to be able to pull that together. There's additional integration that has to happen to make that work. Is that a level of complexity that's beyond the range of a typical enterprise? And I think for you in terms of the enterprise side of things, that's really the focus of where a lot of the energy connection has to get done. What's that level of integration and orchestration that you see on the enterprise side of? Has the enterprise gotten smart enough more broadly? And again with whatever those definitions of enterprise you want to head towards? Sure. I think in general one of the things that I see in enterprises is enterprises have a tendency to look for more complete, more packaged solutions. It's not going to be a platform early. It isn't there. Part of it it may be because of scale. It may be resources. It may be focused. And so it's actually not surprising to me that you saw enterprises go towards VMware before OpenStack where you saw some of the large operators, some of the web scale guys go do that. We see the same thing in SDN and in NFV which to the extent that some of the other people here on my panel can come in and say I'm in and deliver a solution to it. That's very attractive. And compare that to, hey, there's this broad community out there. Whether it's OpenStack, OpenDaylight, whatever you want to tackle, whatever you want to look at it, and hey, go get educated, go find out. There's a wonderful community of people who will support you, but just know that you need to do some level of self-support. That's a lot harder for an enterprise side. So I think we're going to see the same thing here that we see with any kind of technology. At first, people need to invent the technology and so you get a lot of people who create a technology and just deliver it and make it work. After a while what you find is there are many different flavors of it and there's a cost to having a whole set of technologies you can't test across all of them. So it becomes a desire for a standard. It either comes through standardization and there's one way of doing things and people align to that or it comes around a standard platform that everybody is sitting here thinking. I'm sitting here because I believe that we need a Linux for networking. Whether you're looking at NFV or SDN, we need a common platform and that a lot of these things right now that are integrated N2N solutions will simply become applications that run into the platform. That's certainly true around OpenStack on the compute side. I certainly believe that it's necessary on the networking side also. I want to actually dig into that in a little more detail but I also want to make sure that we leave time for questions. So if you have questions if anyone has a question at this point we have microphones over on the side if you've got but let's dig into some of this but if things are coming up keep those in mind is there a point at which these requirements start to converge. Anila you were talking about that sort of shift of some of those capabilities moving into applications sitting on top of those platforms but those sorts of transitions are still a ways off where does that fit and actually Scott I wanted to sort of get some of your perspectives on. Yeah I mean the trickle down NFE is a really good way to look at it I think the requirements are still kind of being understood in the TOCO space but they're the ones that are investing heavily in developing those requirements and the solutions I sat in a BOF session yesterday of the OP NFE and OpenStack teams talking about cross-collaboration and cross-team participation I think that drives some development into OpenStack and OpenStack becomes a more consumable platform for these things and maybe some of the things we were talking about here become an inherent part of the Linux for networking or they become an inherent part of OpenStack and then whether or not the consumer intended to deploy NFE there is a component in OpenStack that they're just going to use automatically so maybe that's where these things start to come in to be used but yeah it's as with any vendor or solution developer in this space it's use case driven and so far we're not totally understanding all those use cases we think there are a few firewall and load balancer and things like that for sure but yeah I guess my standpoint is that it may be too early to be pushing for convergence or looking for convergence that we need to actually explore the use cases a lot more and it's okay if we have several projects that have some similarity in terms of how they're managing some virtualized function the I think Tacker is very interesting because it has taken that kind of general approach of being a platform for any network function and it will be interesting to see how folks can actually take that and use it and it's better by the way I'll mention again Intel's open security controller is an open platform so any IPS can go into that and that's already there for IPSs now for those other use cases it'll be interesting to see if people can take Tacker and modify it for their use case and I was going to say that it's better than what's happening before we were having conversations with vendors often where we'd say they'd ask okay so your SDN overlay base that's great can you do service chaining so that we can service chain the appliance okay yes great we can can you do bump in the wire or routed we talk about that okay so who's going to do the orchestration and the vendor says well we build this network function that's our specialization we don't build orchestration and likewise from our standpoint we build software defined networking we want to move packets we want to provide connectivity and policy and the orchestration of VMs or containers that's not our problem we do that maybe for BGP right I have to move my BGP containers around I might have a VPN appliance I move around maybe a load balancer but for the more complex service graphs whose semantics I don't understand that's something that and therefore the customer would have to build a set of scripts that were just useful for their use case not really shareable the ties into their integration points somewhere around the orchestration in terms of that orchestration integration what are your thoughts about where they fit and what those integration points are for that broader orchestration environment well it I would agree with Pino that's definitely where the conversations are going for some of these specific functions right so from a new watch perspective we can do that same type of service chaining function but then it gets into alright that's great you've got that part who's the vendor that has the other part right and then where those two meet I think there was at least early on some hope that OpenStack would be that place where things would meet and I think there's still potential for that to happen or it's just going to be the fact that we're an open solution with open APIs and the service function vendor has the same thing and customers are becoming more savvy the skill sets are changing so that they are able to look at that and go yeah I can make those two talk together that's fine I'll take that on so I think there's a spectrum I would certainly hope that it moves towards something like OpenStack or some of these other platforms that would standardize it a bit more so that it is easier for vendors to interoperate it's easier from a support perspective there's a lot that's technically possible from an enterprise customer perspective the question is who's going to support it when something goes bang in the middle of the night who do I call, how do I fix it who's on the hook for it there's a lot about that job that's about risk mitigation so it's important to have frameworks to do this and I think there's certainly some potential within OpenStack to be that or some of the projects inside of it that's very interesting we had a question I was this is Naden from Red Hat had a question about the where we stand from a global perspective so yes AT&T Verizon was mentioned I was also interested in the trip to China and that perspective but overall from EMEA and different countries around the world what's the, where do we stand are enterprises out there beyond North America how ready are they very interested in the panel's perspective on that I'll start with some data from an Open Daylight perspective and this crosses SDN and NFE but when we went and we looked we found that there's almost even usage of Open Daylight across three of the five major regions in the world so if you look at the Americas certainly like you said there's a lot of usage however I see Asia and Europe roughly about the same Europe tends to be slightly smaller right now Asia's been it's probably the fastest growing the two areas you just see the least is actually in Africa and in South America so I would say in general that's what you're saying now I will point out a couple other things in the United States we tend to have larger organizations so on average I tend to see in the US we tend to be a little bit ahead and this is true across a lot of areas of technology in China you have a lot more green field so one of the advantages that you're getting in China especially in other parts of Asia is because you're having so much build out right you're going from 100 million users to 800 million users within a couple of years that allows you potentially to do some new things and you've got you know you walk around Bidu the average age is something like 24 you walk around AT&T and it's not 24 and so John Donovan stands up on stage and says hey we're going to virtualize 75% a big part of him saying that is not for the outside audience it's for the inside audience to get them to change and move in a place like Bidu five years ago they were in high school most of those kids and so learning something new is what they do every day anyway and so I I think there are some of those differences to that green field I take a look at a lot of our research absolutely mirroring the same things but that sophistication of orchestration and who can actually start to consume that also has got a really broad driver you get western Europe and the Nordic countries that are starting to do a lot of adoption have already done a lot of orchestration adoption just simply because they've invested early on not so much in North America and green field opportunities China leading but throughout Asia we have another question we want to make sure we have time for Thanks Eric Scott Fulton from the news stack one of the things I've heard more than once at the conference here and I've heard it again in another form here is the theme that if you don't have homogeneity in the platform you can't scale it Neela has talked about asking the question what kind of enterprise are you in order to be able to determine what fits for your enterprise and how NFV applies to you well if we end up with the situation where we have NFV and other characteristics applying to different enterprises in different levels or different respects then don't we create a situation where small businesses cannot suddenly become Baidu if you will they're not homogenous enough to be able to scale up when it's ready for them to scale up because their resources are rooted to a different way of business and I think to Neela's point you wind up having a set of different consumption models for those smaller businesses you wind up having integrated platform capabilities they're going to handle that eventually for a lot of those organizations it winds up being somebody else's capability through a service provider in some form or fashion it's when you start moving up the stack that you now change what those consumption models are and I think the everybody's been talking about which of those pieces of the puzzle individual types of organizations are actually able to absorb who's able to pick up tacker today do all the stitching necessary to make that work you've got to be pretty high functioning in terms of the capabilities that exist with other different options there are a whole bunch of different choices but this sounds like something I'm going to give each of you a last shot at what that looks like I was preparing to answer the previous question let me just add really quickly about the point of scaling only if you have homogeneity so there are various layers at which you can have homogeneity and it's not always the case that an enterprise wants homogeneity because they're trying out lots of use cases and sometimes they need to build out to learn what the right platform is we're seeing a bunch of experimentation with container orchestration models so I think we see even in the same enterprise multiple platforms heterogeneity and then as much as it's painful eventually they try to make those converge as they need to over time because of the efficiencies they need to grow so maybe homogeneity is a property that you end up picking up as you scale and not necessarily having to start out with when you're experimenting to your point about your question about consumption models how about I pass it and I'll sounds good does anyone have a ready answer oh sure oh sorry you're blinded by the light so please sir question there's been a lot of talk about micro segmentation which is the act of inserting virtual firewalls or routers between different components of an application so do you see that taking off do you see end user demand as opposed to vendors is talking about it and can you talk more about that please I mean this is one of the really great use cases that's driving virtualizing those network functions right as my security perimeter changes I by necessity have to apply security policy at a different level and that's what gets us into micro segmentation micro services drive micro segmentation requirements and so yeah a lot of development around where I apply and how I apply security policy certainly drives products and solutions on to the market that answer that so yeah I mean that's definitely one of the leading things it's not just firewalls a service it's how I actually apply that security policy and it could be a firewall or it could be in another way or a combination of ACLs and firewalls and other policies and actually I think you're touching on what is a good point of distinction there which is what is that function which would be sort of higher order processing like a firewall versus just raw segmentation which is access control and hard limits that don't necessarily have any sentience about the traffic it's going through micro segmentation absolutely critical but something that you can do through partitioning of the environment as opposed to higher level functions that have to manage complex conversations or other capabilities beyond that but I wanted to actually start by answering the question that you asked which is around consumption models and for a second I'm going to change hats as executive director I'm also a CIO and I faced this question in that we use a lot of infrastructure and in fact we started building our own open stack cloud at the Linux foundation and at some point I found we like many CIOs in the world my infrastructure wasn't meeting the needs of the business in my case I had developers were creating new tests and it took a lot of time so I went to rack space and I said hey rack space can you take that problem off off my plate initially went to the public cloud now we're on their private cloud and I think you know what this illustrates for me is from a consumption model you really have to ask yourself this question is my infrastructure is my network critical to my business if it is critical and strategic and you must own it then you must think about how am I going to go towards a platform strategy it's fine in the short term to go in and buy point solutions to solve point problems but you incur integration debt when you go there and so you either pay for it up front or you pay for it long term and so for those for whom it's strategic yes you want to get engaged in platform if the platform's not quite ready for your needs it's a strategic investment you invest a little bit you understand that point solutions are short term options and either they will work onto your chosen platform or be replaced within three to five years so that's one side of it at the other extreme you talked about the small company I think the right answer in most cases is go find someone to take the problem off your plate so that may be Amazon Web Services it may be AT&T it may be Verizon it may be Rackspace right and it's interesting it's not that open daylight of the Linux foundation does not have technical talent we certainly do I would just rather have my staff be thinking about other things rather than getting open stack up and running and adding more adding more memory to scale it out I'm happy to write a check to take that off my plate even if we're talking hundreds of thousands of dollars a year so that I can focus my resources on really delivering the new continuous integration tests for my developers because I'm not in the business of running infrastructure I'm in the business of actually creating infrastructure in my case and so I really think that there are sort of those three levels you can either on one extreme try and manage it yourself and it better be strategic you can go buy a set of point solutions from people and probably it's an in between solution or you can go outsource someone for whom it is strategic and then really hold them to are you meeting my business needs and so that's my view on where I think consumption models are going and Jared we will leave you with the final word alright in between you guys and lunch I would totally agree I mean I think we've seen this with a number of other technologies as well right there's kind of a bell curve there are people at the one end that have the technical talent it things are strategic they're the folks on the other end that need to focus resources in other places and so they'll outsource so I totally agree with that I think that that's where it's going to go what it looks like in each of those areas I think there's room for growth and I would say we're still probably pretty early on in what those consumption models really look like I think they're going to evolve in each of those categories but and but I'm encouraged by all the work that's being done the more that I think it can be done out in the open in open source communities and open source frameworks I think does encourage folks to get involved and to provide input and feedback and I think that's really important well watch this space see how things develop I disagree but there's no time the what? I disagree no no I yup we'll take it outside I would just say that if people tolerate me for just one second I don't agree with this approach that you can just say just take the problem off my hands I mean you often are trying to move faster than the public cloud provider and you'd be at the public cloud provider if you could put your data there or if they had the the requirements the features that you're looking for so often the customers are looking for something that's not being done yet the platform isn't really mature enough so they have to get they have to commit to a platform that is still changing very rapidly so ideally it's a good business logic but in practice today people are having to cobble together solutions with platforms that are great and have made tremendous strides but actually need a lot more maturity well with that we will leave it once again to join me in thanking our panelists thank you very much for attending today