 Okay, so you can hear me. I guess we can start Thank you for joining. I'm actually very pleased to see the turnout this afternoon. It's fantastic to see the level of interest here Over the next hopefully 40 minutes or so if I still have them we'll be covering The top 10 things net app has learned about being in the open-stack marketplace and just to caveat up front This is not a vendor pitch. We're not trying to sell a net app hardware software here It's literally our observations as developers as participants and contributors to this particular community So brief bio on who I am I've been a net app about 15 years or so now it'll be 15 years at the end of the month I've seen a lot of change within the industry the storage industry the data manager industry in particular I've also been privileged to actually witness our participation in various open source communities various Industry associations and standards efforts. This is a very brief bio of the particular Industry association standards and open source communities. I've been personally involved in just so you know net app actually Maintains the entire network Portion of the Linux kernel. We actually fund that development entirely as our contribution to the open source community as well as Ability for us to leverage and introduce new technologies. We're very actively involved in the BSD community as well If you know anything about our core technology data on tap You'll know that it's now based entirely on the BSD kernel Subsets at least of the BSD kernel and of course we're involved in Hadoop and the open stack community as well So this is just a brief bio on what I've been doing one of the things that I'm not gonna touch directly on here But it's very very relevant One of the first standards. I believe first cloud standards to exist I was able to help introduce to the marketplace That's something called CDMI cloud data management interface And it's an abstraction a superset of both Swift and S3 restful interface very very well thought out and well defined And that's something to have questions after the session about how it might apply to this community here How it's being adopted globally. I'm happy to answer that as well So here's what I do hope to accomplish today. It's really all about perspective. It's about learning what we've observed Over the past I'd say 18 months or so at least that I recall us being very very actively involved and dates back we to Diablo's as I'm finding out We've been involved in this marketplace from a number of perspectives So I'm gonna reflect the customer perspective Then the storage and infrastructure industry perspective We'll talk about some of the workloads and actually some of the development workflows in particular The competitive aspect here is really what choice we're seeing customers make with regards to all things cloud Where they directly apply to open stack or whether they apply to the broader cloud marketplace such as it is We'll talk about the thriving ecosystem that's evolving within the open stack community as well So without further ado, there's actually a very very good blog that was done by one of the VMware engineers a little while ago talking about really the mentality of The cloud or Adi or the cloud communities and this is something that I resonated with very very directly when I read it about Six or nine months ago and it was published and that's the notion of if not specifically geographically but a state of mind That exists within the technical community here. We're at a conference here It's already West Coast based all of us are you know highly technically involved people many of us are developers We have to continuously remind ourselves and this happens to be fortunately every time I meet with a customer particularly outside of one of the key High-tech hubs whether it's in Middle America or Middle Earth so to speak whether it's Overseas in a certain parts of Europe or a pack Asia pack and so forth Continuously reminded that the level of risk tolerance the level of early adopts if you will is so much lower It's almost night and day outside of the eco You know outside of the community that we're directly involved in and the self-fulfilling loop and feedback loop that we have here That we really have to appreciate how adoption Really really rapid adoption that we're accustomed to here and some of these really rapid development cycles is not the norm Outside of our particular demographic and that's a very very important thing to appreciate as we actually try and make Open-stack and mainstream technology make it as ubiquitous as the word windows to the average layperson And associated with cloud Now before we actually look outside I felt it was very instructive to look within what does a Typical and I actually categorize net app now as roughly a seven billion dollar 13,000 person company were the classic mid cap company nowadays so looking within to find out how net app site tea department actually Segments their cloud budget allocation their cloud resource staffing and of course most importantly the cloud services They offer me as an IT user inside the company we quickly see that one of the first things that We were focusing on is Net app IT I should say because I'm a part of that up it but that net up it focuses on is Being able to rapidly provision manage and reallocate Existing third-party applications effectively binaries not source code So not so much a developer mindset very much an operational mindset of existing technology This slide is sort of drawn to scale. So that consumes about 20 25 percent by far When you talk into a business person When you talk into a business unit lead or he actually average user inside an enterprise by far The biggest association of cloud is with softwares of service more than infrastructure and way more than Platform of the service and of course by far the leading shining example of that is our friend beanie off company salesforce.com and Essentially what we're seeing here is the fact that there's a large sales force dominated cloud That a lot of IT budgets a lot of IT staff is associated with managing with governing with setting policies for and there's a whole bunch Of somewhat random other third-party applications. So at net app. We use workday. We use service now We're using a couple of the marketing ones Mercado radion 6 list was on and on and of course some pretty familiar names at the bottom There were the guards to you know high volume high profile collaborative software the service technologies such as Twitter Facebook LinkedIn or so forth and again This is drawn more or less proportionately to to budget and staffing scale. So this is really what's left over when it comes to really flexible agile infrastructure clouds and As you probably know, there's certainly no default assumption that all of this is going to be open-stat There's a huge proponent in the company that you know with an IT that wants it to be for some obvious reasons that we all share in Common here, but this is a competitive landscape There's actually a bunch of different categories involved and this is really how We now focusing outside of IT actually and on the engineering side of the house Particularly in my group the office of the CTO. This is how we view the marketplace we actually view it from a business perspective on through the technical and you'll actually find the flow of this presentation is very much around a Business view and a strategic view and I'll hone in more narrowly down on a technical perspective as we get to number one So from an industry perspective if you look at sort of the top and bottom of the slide first you'll see very very different business model approaches towards Industry delivery of cloud services and cloud infrastructures at the very top to buy and operate categories all about using the power of your balance sheet Using a lot of actual non-technical assets and non-technical people you might have in your company to buy Third-party or implement open-source technology if you will but primarily by commercial technology with all the commercial associations around sales support Installation support posts in an installation support and so forth professional services That whole sales model is built into that nature of cloud deployment So think of it as third-party commercial cloud technologies at the very bottom You've got the polar opposite effect of folks that Essentially have data as their business Data doesn't support their business. It is their business almost an entirety and as a result the business model Means that you can justify a lot more investment in actually hiring the engineers and Hiring the expertise and developing nowadays literally from the silicon on up all the technology You need custom tailored custom fit be spoke if you're from overseas for your specific set of requirements So it's almost a polar opposite where there is very little if any Third-party technology at some of these companies is very little at any third-party Commercial technology some of these companies a lot of it is based on open source Historically software and with the open compute movement even some open source hardware right now Back place for any of you that know it is a really interesting backup as a service provider And they're one of the first actually open source their hardware disc shelf specs So a lot of that a lot of those themes are really filtering into the business models And in the middle is I think where we fit here particularly at this conference In the middle is where we certainly are not looking at Owning all of the IP and owning all of the engineers that deliver these custom Hardware and software layers, but actually forming an open community a rebel alliance if you will in Star Trek parlance and Being able to you know Leverage the power of the community and primarily the enthusiasm of that expertise inside the community and come up with an alternative That tries to deliver the best of both worlds tries to have a lot of expertise and innovation in a community context But not necessarily have that expertise bound to one commercial company and Delivering the technology in a very very open manner Where the classic model of course is to profit from the professional services or the support? But not not at all from the licensing of that technology and Each of these business models has very direct again implications on a specific research and development strategy of these category of companies a Very popular blog. I guess it was a Randy bias But some others have picked up on this concept. It was written around the difference in this community between your pets and cattle Meaning and from an infrastructure perspective You know Can you actually name and identify and hug a particular server a particular rack in your data center? Or are they so ubiquitous and so anonymous and so disposable and so replaceable That you've got an architecture mentality where you can effectively just roll them in roll them out your application Availability isn't impacted by any particular outage of a hardware component or even sometimes if you architect really well an entire rack That's a difference. We're talking about here And the way I see it from a user perspective from a developer perspective From an infrastructure operational perspective, this is really not about polar opposites You'll find this scene throughout a lot of my talks including this one. It's not about a strict either or It's very much about the spectrum and where do you fall on that spectrum and Specifically around consumption models. How do you want to consume infrastructure in your open-stack design? What we've seen in the marketplace continues to surprise me and at the same time reassure me We actually see already these polar opposites You're actually seeing a bunch of raspberry pie motherboards over there on the right and a category of converged Infrastructure, which is something that you might know brands like flex pod V block HP Forget what they call theirs IBM's pure systems and so forth These models are both succeeding wildly in the marketplace nowhere near at the expense of each other Which means at least at the polar ends of the spectrum. They're solid, you know market-driven demand for both And what's interesting is how the spectrum itself is widening out? so in our case particularly at net app, what we're seeing is the ability now to enable our partners our customers To our developers inside and outside the company to consume our technology using a mult, you know a variety of different interfaces or flavors Over on the left-hand side you can start with these fully converged fully integrated extremely engineered racks You can move gently over to say you know what I do want to do sort of a best of breed piecemeal approach I don't want the entire thing fully converged for me I want to pick and choose the best storage layer the best file system layer the best network layer the best server The layer the best hypervisor layer so be able to decompose that in our case That really was where the business started our classic appliance model You actually see people decomposing even the appliance on saying you know what I've already made investments in someone else's disc drives Or I can buy my discs way cheaper than you can provide them or I value my discs a lot less than you value the price And you give me so why don't you decouple the discs from the actual controller hardware or software that runs your technology? And that leads to a full Software-only mode some people are I thinking correctly referred to this as software-defined storage now the latest buzzword in the industry But effectively a software-only implementation of technology we deliver that has no hardware bundled in no discs bundled in and depending on certain testing and hardware Certification lists you can essentially mix and match any controller and disc you want and all the way to just a virtual instance Which can be a hypervisor based instance or a cloud-based instance of a storage volume or a storage Array of virtual storage array in the cloud that you can manage You can instantiate as easily and as elastically as you can any other resource in the cloud You can manage as easily as you can any other resource in the cloud And of course deep revision or eliminate as well as any other resource in the cloud But benefit from that eat that that strong desire for elasticity that exists there So the the ability to deliver these multiple consumption models is I think the key here more than One particular religious flavor of all open-source commodity versus fully engineered commercial rack Hybrid clouds come up a lot And I've always been reluctant to talk about the cloud bursting scenarios Which people love to theoretically discuss, but I've seen very very few examples of an actual practice Particularly compute and storage bursting particularly storage bursting is notoriously difficult If you've actually seen Dave McCrory talk or seen his data gravity website or Understand the concept of data gravity you'll understand why literally it's hard data has gravity has mass Compared to compute. It's so much harder to move data around that it's easier to bring the compute to the data than it is to move the data to some commercially convenient compute facility So as a result the kind of hybrid cloud that we're seeing most popularly deployed in the marketplace is in some senses a recap or Back to the future moment of the old sand and nasties where Data center managers realized for flexibility and for maximum resource utilization Decoupling and separating compute from storage infrastructure and being able to network the two somehow Gave you tremendous flexibility and dramatically increased either asset utilization or enabled Dramatic it was called server consolidation or file server consolidation in the old days, you know dramatic reduction of unnecessary resources physical Capitalized resources. We're seeing the exact same thing now play out in the cloud era where and it's not a cloud era talk about who do this is the cloud era and We're seeing it play out in That people are realizing for regulatory reasons. I'll get to in a second or for just personal emotional reasons or financial reasons They need to own the data literally They need to actually have the data sitting on their capitalized storage systems But they still want to benefit from the tremendous compute elasticity available either and on you know Their own infrastructure clouds managed by open stack or third-party cloud the open stack based or Amazon or other So the notion of actually parking your data in colo facilities right next to cloud data centers and benefiting from very very high speed links Between a public hyperscale cloud and that colo facility so that you can access the data as if it were in the cloud But it's not that's proving to be the most functional usable and popular Hybrid cloud scenario out there. We actually split the compute from the storage and that's the hybrid boundary So that's not what I expected years ago when I first got into the cloud business But it is what's playing out right now That's not to say this is the ultimate vision of where hybrid clouds go in the future But this is reality today and it's proving it's exceeding my expectations in terms of popularity So actually I guess I didn't touch on some of the regulations, but this this kind of applies as well The continuous integration continuous integration discussion versus, you know having more phase or staged adoption This comes up a lot in the developer centric communities particularly more dev than ops Where there's a strong desire to just push the code out push it out get it visible get it used get it used Debug it basically in production and I think for a large class of Technology we use today of 80% maybe of all the apps on our phones that actually makes a lot of sense You know highly collaborative apps apps that don't actually own the physical resiliency of the data But just you know our consumers of the data Those apps are perfect perfect. I think candidates for this kind of technology and again this comes back to the notion of is your Is your business all about data is your business data effectively? We can do this or does your business use data in support of its other functions And so this is why I think you know the analogy of eating raw sushi or the sushi fail Well where you can tolerate some outages at least the availability of the data you certainly can't tolerate data loss, but availability of the data performance Expectations response time variability is tolerable and so forth versus having really a fine cooked meal and If you want to think of you know many many categories these six or so are just off the top of my head and various Customer verticals that I talked to of where it's completely unacceptable to debug in production and to push data out Very very rapidly think of you or one of your loved ones in surgery In open heart surgery Do you want any I mean any of the software associated with the success of that surgery using it? Continuous integration model as it is today. I can tell you I certainly don't Think of you know how we all got here most of us don't live here Whether it's airplanes and some of the control embedded systems in the aircraft itself or whether it's the air traffic systems And do you want rapid continuous integration integration in all of these environments where you know literally it's the difference between life and death? No, I definitely want a much more planned phased approach same thing for you know automotive control systems If you buy a car today or over the last year or so there it is a you know literally frightening amount of lines of code in your car that are controlling your braking systems controlling your airbag when they should or shouldn't be deployed Controlling your lighting controlling, you know, certainly if you take advantage of some of the more sophisticated You know cruise controls now controlling the distance between you and the car ahead of you and some cars behind you and so forth I'm not even talking about the Google car that we see on the one-on-one all the time in the valley, right? These are things you and I can buy outside of Mountain View and we are buying and we are driving right now Millions upon tens of millions hundreds of millions of lines of source code there I don't want that in fact. I don't think I can have that continuously integrated same thing for military systems Whether they're drones or tanks or battleships or what have you you know? Let's make sure we debug those nuclear power plants I'd like that software to be very very well thought out and tested and the one that actually get beat up on every day Maybe just because it's a vocal community or maybe because it's New Yorkers That's may like to let you know how they feel is in Wall Street They suffer from and you know the IT guys really do suffer the analyst benefit But the IT guys really suffer from an overwhelming amount of regulation You know post Lehman Monday post the dot bomb crash and all these other financial crashes we've had And they literally can't they haven't figured out a model for rapid continuous integration In you know in their business So they even though they'd like to be able to offer you more services and more weird financial You know constructs to be able to buy artificially and get into another crisis. They just can't there's too many regulations right now preventing from doing that All right, so switching gears now to open stack specifically one of the other things we've observed is many many Customers looking that are non developers particularly that are not directly involved in an open stack project So this is the majority now of people in in the open-stack world if you will on the customer side are Very very confused as to what storage really means in the open-stack world Because out of the gate out of the bat There's automatically in this dashboard as you can tell it if you develop actively is already a couple of versions old But it still illustrates the point There's two different kinds My favorite joke in a blues brother's movie. There's two different kinds of music country and western Same same thing and in storage here. There's really two different kinds of storage, even though it's all open-stack There's the compute, you know application oriented storage and support of the compute side Particularly all the instances that we have and we know it we know it as volumes today The sender project in the open-stack world and then there's a container or object view Which is a vastly vastly different technology vastly different interface and more importantly one of things we keep observing is that Whereas you just merely stop at the abstraction in the interface for how volumes are created managed and deprovisioned and associated with no instances and so forth The Swift project is all about how you implement the technology as well as the interfaces the object interfaces to your containers And we think sometimes that's a little bit too much of a tight coupling there That's too little bit too symbiotic and perhaps not healthy for the future of open-stack and even swift as a whole So this is you know if you're a dickens fan as well This is really very much today in the open-stack world at tail of two cities You've got the revolution going on a lot of Rapid development and emerging technology in the open-stack in the open swift the ring space You've got relative maturity Over on the compute side when it comes to a very well-defined model with sender for how you create volumes how associate volumes of Instances and so forth and I think the deployment and adoption model certainly from the feedback customer feedback we get reflects that So let's take a pause here Take a deep breath because this is the busiest slide by far and I apologize in advance for the amount of animation in this Slide, but it's really the only way to get the point across back to the theme of diversity Customers have a choice and specifically the community we care most about here Developers all over the technology sector have choice and this is really the I think the flora fauna if you will of the entire Diverse ecosystem that they have available to them a lot of it as I mentioned earlier includes commercial options as well as Proprietary commercial service options and not just open software not just free services Although curiously enough the honey pop model here is that a lot of these developer Development environments these IDEs are often offered up as a freemium model Or a very very low cost of entry model But there's again a wide diversity at the top in terms of application Development I'm not talking about deployment yet, but application development platforms And in fact, you know I kind of wish in hindsight that I was about 20 or 30 years older for a number younger for a number of reasons But the first one probably the geek side of me is I just love IDs today And the ability to develop in the cloud and not even have to buy the hardware or worry about keeping up with the latest You know PC or laptop processor speed for Christmas of the IDE That's just a tremendous luxury that people I think that didn't live in the old days like me that the caveman days of computing You really shouldn't take it for granted. It's just amazingly wonderful time to be a developer and these services reflect that But it is a honey pop model remember none of these guys are in there for for charity Here's where I would be scared to death if this was thrust upon me today at the data layer the breadth and depth now, but especially the breadth of options for persisting data is Maybe not for young developers that grew up, you know, I don't think a lot of developers are digital natives yet But they're close But for for me in my era, you know when I learned relational and set theory was pounded into me in university This would scare the crap out of me. There's just too much choice here I'm familiar with you know, literally the polar ends of this. I'm familiar with the relational model when I grew up and Just, you know, simple file block interfaces to persisting data But now we've got tons of options for column stores native document stores the uber popular key value stores And just object interfaces rich object face interfaces of all kinds It's a very very rich batch services now that are led by the Hadoop community in particular and the thing about this is It's the new reality Some of the logos underneath each one of these categories may consolidate Yeah, just a natural effective maturing in maturing industries, but what will not go away is literally those categories themselves So we are in the world now. We're literally a polyglot data world where we do have to be very very comfortable Familiar with these various different kinds of data layers data services layers and the open stack community has to I don't I haven't seen it yet really has to embrace all of them If you really wants to become truly mainstream and want to have higher and higher orders of value To the developers in this world because developers using all of these and we don't want to have an infrastructure layer That's bound to just a subset Again the infrastructure layer itself is getting very very interesting right now because now we're talking about not app development but app deployment and It's very seductive once you've developed one of these applications with a very sexy gooey and IDE to just click You know deploy and have it hosted on the same platform you developed it on That also has of course a lot of long-term consequences It's a it can be a higher cost of ownership higher cost of delivery And of course the ubiquitous concern around Supplier lock-in services lock-in, you know service provider lock-in vendor lock-in call it what you will So being able to abstract as much as possible Across the data services across the infrastructure services And this is what I will put in my shameless plug for CDMI the cloud data management interfaces at the data layer being able to abstract particularly for object and containers between any one particular kind of Restful interface whether it's s3 swift or other that comes along whether it's a service or an actual array a product That abstraction can be incredibly powerful because at the very least it has you know, I Would say two-fold major benefit two high-level benefits One of course is if you're an application developer you code to the abstracted API and Now the data or the namespaces you create in fact can either move seamlessly between back-end storage systems or providers or both Or of course you can create hybrid namespaces effortlessly By essentially just mapping the namespace across an intelligent abstraction layer They can abstract multiple back-end clouds multiple back-end storage systems or any combination thereof So as a developer it's a real wonderland to be able to create these rich Hybrid namespaces and not have to worry about the physical location of your data anymore that decoupling the physical from logical It's incredibly powerful and useful and in fact what I've seen is the only way to manage the data at scale That we're managing right now like quote-unquote big data. We're all dealing with now from an operations perspective this tremendous benefit as well Because again at the operations level you're the one that has actually bind these applications to something that runs them And if you physically and very tightly bind these applications to Specific infrastructure specific vendors or specific service providers You're just locked in again, and you have very little power very little leverage And it might be a very easy decision Oh, you know initially but a very very painful and expensive decision over time So abstraction layers across multiple storage systems storage services Cloud providers is a big theme. I'm seeing as the cloud market overall not just open-stack matures I actually do see these delineated boundaries in terms of customer behavior almost every year in 2013 Keeps coming back and kind of slapping me in the face and all my customer discussions 2013 is a year of paranoia around cloud locking whether it's storage services or service providers as a whole and People are thinking about it now No one's really doing anything about it yet But people are thinking about it and it's a very healthy and important thing to think about because again the two main options right now Are to end up either in a commercial ecosystem like VMwares or a you know very very popular but private proprietary ecosystem like Amazon's and Those can't be the only two options Okay, so Narrowing down to the end here distributed versus centralized control This is another kind of like continuous integration religious topic here And what I find fascinating is observing what's happening in real time right now around the different Mindsets being applied to the exact same collective open-stack project You've got quantum taking advantage of all the hype and actually practical benefits of software defined networking That are centralizing control in the architecture versus having a very very distributed network control in the past But on the storage side, you've got the exact opposite It's kind of the Wild West cowboy days of everyone wanting fully distributed everything when it comes to storage And not really realizing some of the dangers and the ramifications of that So it's really fascinating that the networking world is coming full circle. And of course, it's inevitable That the storage world will come full circle and because as with many things in computer science This is we're having a bit of gray hair or you know, no hair all in my case right now But having that experience is extremely beneficial Because as we can date back to in 1962 here Paul Barron and some of the research he did around distributed networks and distributed network architectures There's always a tension between centralized models and you know, it never looks like a ring in real world That's what it looks like in real world distributed models and a hybrid sort of decentralized approach We're finding at the storage layer has tremendous benefits. And here's specifically why Resiliency This is sort of the one of the number one thing that comes up in a storage discussion is how can we implement low-cost resiliency and stick to you know avoid vendor lock in use commodity open source hardware and software components and the reality is you can't But if you take a look at the most mature implementations of ring storage technology Amazon's Dynamo and Amazon's S3 that's based on that same technology How many copies do you think and I don't think I'm revealing proprietary information here But I do know the answer. I'm gonna see if anyone in the audience knows how many physical copies Do you think S3 deploys to deliver the very impressive level of resiliency that they do deliver right now? Anyone willing to take a stab Hearing two and three and I'm gonna be like an auctioneer any higher bids Much closer. I heard 13 in the front. It's actually you overshot it. It's 11 So forget about the steer theory of three copies in your safe Okay in practice in the real world 11 copies physical copies of your data are required to deliver resiliency at scale It is fundamentally broken unless you run an interesting business model that Amazon runs for all sorts of reasons Three logical trust me. It's I'm talking about physical copies here And we have a bit of time at the end so we can certainly have a discussion about it offline The reason why we prefer the decentralized approach in the middle here is that we don't think a ratio coding versus raid Which are by the way different flavors of the same math It's basically parody versus linear algebra, but it's fundamentally a similar math. We don't think this is an either or discussion We have done whether it's across all of that app engineering or specifically in the office of the CTO with university academic partnerships with Internal research projects and specific models that we've created. We've done a tremendous amount of research in this topic And our final conclusion is that the closer you get to the disc the closer you have these really rapid interconnects these back High-speed back planes passive or active whether they're PCI or or much more sophisticated than that the closer you get to the disc The better it is to use parody based raid to recover efficiently cost-effectively from these Physical outages these disc level outages the further you get from a disc so particularly You know groups of shelves or nodes if you will in most of these common designs and you want to be able to protect data Efficiently instead of 11 copies way less than that less than two copies to do it intelligently If you want to do that across nodes particularly at a distance so they're getting outside of PCI or memory bus territory and into Ethernet territory and wide area networking technology. That's where a ratio code. It makes a ton of sense Because some of the latencies Inherent and some of the compute overhead and during the reconstructs are masked by just a general latency of the interconnects themselves So a best of both worlds and I'm not going to pre-announce anything here But future technologies from that app are going to include best of both worlds here I'm going to include both of these concepts and very very widely scalable Very low-cost hyper-stale storage infrastructure So that was 10, but of course we're all engineers here. We all love spinal tap so this list goes to 11 and This is a perfect example of us contributing something back specifically to the open stack community Rob here on the side, and he'll be happy to take questions later on Along with Ben who I think is in the audience as well There you are I have already published a proposal including a blueprint for a shares service a file sharing service so aside from some of the Strong technical benefits for considering file base as opposed to just purely block-based services for Cinder or you know other constructs as the case may be over time Just take a look at the market You know IDC is tracked the latest full year that they have data for 2012 of all the storage That's not in our laptops not on our phones, but actual managed storage in data centers Two-thirds of it are not block interface So I know that the community here is all about rapid development and iteration, but why on earth Would you want to ignore two-thirds of the market? I can't think of a single reason You got to embrace this market if we want open stack to become mainstream Which means embracing all of it not just one third of it So I think regardless of whether this particular proposal wins out or not and it's been very well thought out And it's 18 months two years old at least right now. So there's a lot of iteration built into it There is a desperate need For more than just a block or object storage service in open stack And again the market has spoken the market wants this technology and there's very very practical benefits again This is back to the future for me 10 15 years ago sand and ass days, you know if you're able to centralize control Either at a file system level or even an object level and be able to offer a POSIX based file interface as a volume That multiple instances can simultaneously share as opposed to just be bound one to one to there's huge scaling benefits There is huge efficiencies of scaling benefits there and actual data sharing possibilities Within the instances of your apps that are incredibly powerful and are being delivered by the way by VMware I'm being I'm not sure so much in Amazon. You can kind of hack it easily, but it's not a native service Yet, but are being delivered already and of course there's dependencies and non-virtualized Non-cloud data centers inside enterprises today on this technology So again as you can tell I'm pretty passionately lobbying for the community as a whole to seriously look at this proposal Comment on it and prove it or say, you know what it's crap. Here's a much better implementation But there is an implementation needed for ultimate mainstream open-stack success in this marketplace And in a couple of minutes I have left I'll kind of wrap on some things beyond the top ten list that after observing the community that we need to look forward to We need to plan for So one of my favorite Futurists my the only futurist really that I follow that I've met But a guy that hopefully you've heard of because Google has employed him now and gosh knows what he's gonna be doing inside Google right now, but Ray Kurzweil if you don't know who he is look him up Google, please He has actually done some fascinating analysis of how technology dating back to the industrial revolution The rate at which technology progresses So a different kind of continuous integration if you will and what he's observed is Fascinating because it's so counterintuitive doubt to how humans feel or think He is observed now. There's no doubt anymore here, right? This is just mathematically plotted and proven that technology progresses Exponentially not linearly as we think as humans So what are the ramifications here? Well as you can read and he actually the model he uses is the power of computing you can hold in your hand One of the things he was famous for in addition to just doing this research and publishing it He actually applied it very early on he delivered for you know hearing impaired people a Handheld device that looks like an iPhone but this predates the iPhone by almost 10 years and held device It was able to actually be able to you know Intonate some of their their words or be able to actually translate some of the things that they were the device was hearing so that they could hear it So this is very very early version of Google translate if you will He delivered that a long long time ago about 12 years ago now And he keeps postulating on what this graph and specifically illustrates is how much power You'll be able to hold in your hand in the shape of an iPhone iPhone style device Smartphone style device not a faddle something even smaller than that And what he says is around if I'm reading this right I keep forgetting the dates around 2050 or so 2040 2050 The collective power of an entire human brain Okay, this is a lot of brain powers of pardon upon a lot of CPU power Will be available in a handheld device in one single handheld device That's how much power you'll have today an entire human brain still can't be modeled across multiple Vast data centers at a Google or a Facebook or an Amazon would own there's still all the synapse is firing Until we get the quantum computing still require an enormous amount of of CPU power by 2080 that handheld device Will be able to have the Collective wisdom and processing power of all of humanity in a handheld device So this is really how rapidly technology is progressing. It's not just about you know Emulating an insect brain or a mouse brain the human brain is it can tell follows very quickly thereafter So we're very early on in the open-stack evolution phase We're still in single letters not double letters yet We haven't even gotten to half of the alphabet and one of the things I've learned being a dead app for 15 years and managing complex software projects is that you've really got a plan ahead and you've really got to start not just architecting with a Very clear view of the future But running parallel development streams if you can a tick-tock model for development not just thinking one release ahead only and If you if you don't think the theory is practical here's some of the science of the actual practical reality One example of why this is so important at the storage layer is that by the end of the decade We will not recognize what storage looks like managed storage look like looks like What we have again from a storage perspective in our hands Android device and blackberry and iPhone what have you is trivial. It's all solid state-based, which is interesting But the revolution that's coming in storage is not about better tapes Which will look still exist at the end of the decade by the way or better discs Which will still exist, but in a shockingly singular format only it's about how memory is becoming more like storage than storage is like memory and how tearing of RAM and tearing of memory and Processing data first copies of data in memory literally is what the future of applications and computing are all about and There's no better time than right now in the open-stack community to start architecting for that Working with industry associations such as SNEA where Intel has a technical working group around non-volatile memory extensions To instruction sets to libraries IO libraries and the whole notion of IO and in memory computing is an oxymoron So you've got to really rethink about how you persist in memory data structures And how you manage in memory data structures across nodes across racks Across data centers because data management never goes away But certainly some of the technologies and the incredibly rapid processing of power and latencies available by Managing data over the memory bus as opposed to over an ethernet connection. That's a really powerful construct That's upon us and by the end of the decade will be you know We predict widely deployed because it'll be commercially affordable So I'm happy to take questions later on about the specific flavors we're talking about here And I guess that is my last slide. So with that I've got about five moments. Thank you There's a microphone here. I think for posterity in the recording We're probably gonna have to ask people either step up there or of course, there's no group questions I'll be happy to entertain them one-on-one Yeah, two quick things about one. I'm a little disappointed you didn't float the idea discussed about renaming open-stack Skynet The proposal is in fact actually put a type code for the file share of service. So have one. Thanks great Thanks for that and the Skynet reference is very true Other times when I give this talk I say, you know is the future of the data center science or science fiction Because literally a lot of the science fiction concepts are becoming science and reality All right, I'm guessing not a lot of people want to go up to the mic So thank you and I'll take questions one-on-one