 Hi and welcome to a special presentation of wikibon whiteboard. We're here at the Dell EMC Executive Briefing Center Happy to have with me Guy's known for many years. We've done a ton of videos. Chad's back. It's been too rough. We're we're getting older, dude Yeah, um, you know we keep getting older, but yeah, something something Titles are irrelevant. Yeah Great well Chad and great point. I what I'm excited to be able to do is kind of you know Take off the jacket and have a little bit more of a fun interaction What you and I love to do is let's talk about technology Let's talk about some of the trends that are going on in the space Harkens back to both you and I back back when I was at EMC. We did it does a whiteboard videos Back then almost a decade ago was the storage protocols You know you did a great series talking about VMware and the choices customers were making back then Yep, NFS ice guzzie and fiber channel. I did one introducing fiber channel over ethernet as the new Potential channel down and I mean tens of thousands of people watch these videos which in our they're a little neck of the woods What was pretty exciting? I I I can't believe two things number one that anyone would ever get into a passionate bar fight over ice guzzie fiber channel or NFS, but people really did And in fact even on this weekend on Twitter There were some people who were arguing jump the classic jumbo frames versus no jumbo frames argument, but Stu you and I are the whiteboard Kings Yeah, up you got the markers already, so Chad, you know what we talked about it 2017 Yeah, so hard to believe so much has changed in the industry I'm not having discussions about storage protocols anymore. It's been many years You know your your group of courses done kind of converge infrastructure, which I think was one of the main drivers to get rid of it I mean, I remember one of the first customers I talked to that had a broad deployment of fiber channel over ethernet. It was a net up customer It wasn't file. It was just like, oh, that's what you're using. He's like, that's what the solution is So I just wanted to use it now the thing that I'd say though is is that while protocols frankly weren't that relevant then And aren't that relevant now? There still is a lot of debate over What's going on in the world of persistence? I always use the word persistence as opposed to storage Because storage makes humans visualize storage arrays And often I get asked we're here at the briefing center. I literally I just talked to a giant customer that's that is there any What's the future of these things? Are they are they dinosaurs? Are we going to keep using them or what's going on? Or is everything going to move to cloud storage? So there's there's still an architectural Discussion over what what's the future of persistence? And I think the reason that that's important is anyone who's watching who's had a bad storage day And I'm sure people are watching have You know that when information goes bye-bye Um, it's bad. It's really bad because It's the one part of the compute network and persistence pyramid where the information is once it's gone. It's gone So it's important to get right Now the question is what does right mean? Yeah, absolutely I mean I remember you know working here at emcee the four letter word that you never wanted here was du dl Yeah, data and availability or data loss that was unacceptable Um, and right when I look at hybrid clouds kind of one of the big discussions We're having in the industry these days two things I say is you need to follow your applications and follow your data And that's going to help you drive those decisions as to what you're doing And what underlying choices you're going to make so let me let me answer the question here really specifically And I get I want to look right at the camera. Is that cool? I'm not trying to go ahead chat So the first thing I want to highlight Is that the mistake is to think about this being an answer where it's all one thing or the other But I want to make a binary statement. That's pretty black and white our point of view Um, and here. This is not just del emc. It's also del technology. It's also vm worse point of view is that Software-defined storage stacks are ready for the majority of x86 workloads by count Now if you unpack that statement it means a lot, right? Um, so first things first is the leader in the external storage array business That's a pretty bold Statement sure right But if you unpack it the first thing is that not all sds's are the same So you have some that are transactional. You have some that are unstructured, you know data optimized Some that are geo distributed object stores. Those are all different forms of sds and then notice how I said Ready for the majority of x86 workloads by count So You know, if you look at the workloads that are out there, we still have unbelievable workload diversity, right? And you find some workloads that require specific functionality in one domain or another a specific performance in one domain or another and While a two-axis chart is probably, uh over trivializing this Uh, it's a good it's a good place to start All right, so let's start their challenge So if I say kind of oversimplifying once again, you know traditional storage arrays and the storage industry which is changing greatly Especially I think everybody understands with the dell emc kind of merger acquisition there Storage industry is going through a lot of changes and therefore there But if I look at how predict storage has traditionally been bought It was bought as storage arrays and it tended to be Higher performance had higher functionality lower performance had lower functionality You had kind of smb mid-range and large. This was you know, kind of, you know, you know, you know, you know, you know Your traditional storage box. Yep, and there were many pieces in there So, you know criticism of bmc would be like, oh, well, you've got four different boxes that makes it up Whether it's you know, your scale out nas or your mid-range or your tier one or things like that and today You know chad, where do the discussions of all flash arrays, you know hyper converged infrastructure You know cloud storage fits into this where do they start eating away traditional storage? So the first thing I'd say is that the biggest thing going on. Where is the bulk of the information going these days? It's not in things that people would think of as storage at all, which is really cloud storage There's a big giant blob here Cloud and what do we mean when we say cloud storage? Well, we mean object stores. They're geo distributed It's actually not necessarily germane whether it's on-premises or off-premises But it's basically something storage that you access vn api, right? uh, you know, you open a Session that's a restful api from the device and you say i'm gonna put something or i'm going to get something in or out This is the single largest consumer of data That exists out there now it's performance criteria is By transactional storage levels very low um, it's aggregate functionality is enormous right, so uh Some of our largest customers have got geo distributed object stores that are measured in the exabytes of size But they don't want it. They don't care about things like um, you know remote replication in the traditional sense they care about geo distribution of buckets and objects they care about uh encryption and replicas, but not a traditional snapshot Yeah, and there's some really interesting use cases, you know talking about that at uh, I was at the amazon reading that show towards the end of 2016 and Lots of people companies that were built to do analytics on kind of the data lakes We're seeing more and more that those are happening in clouds and a lot of that in public clouds But of course still many going on, you know, hosted private, you know, whatever you want to call those clouds so that's doing there and even Talk about the fragmentation big discussion by some of the people looking forward as to what's this whole Sensor iot wave going to do because when you talk about just the sheer number of nodes It's going to be at the edge and therefore it's not going to be at some core, you know public clouds So storage keeps changing and changing it now I would argue my own personal opinion that the iot thing is going to have a combination of edge Aggregation and core and cloud storage models will be one or the other I think you know people always want to turn these debates into one thing or the other and the answer is usually A bit more complex. Yeah, now the other thing that is fascinating is we're seeing a resurgence resurgence of plain old dads Basically discs that are attached to a host that have no magic secret sauce on them At the storage layer whatsoever at that layer, but then the layer up. There's a lot of magic So Are you saying that that type of convert infrastructure or you're separating that out? What i'm pointing out is that there's certain use cases where the customers are basically saying I am going to build in the intelligence into the application layer. So for example, uh, you know When you and I first started to work together, man, the amount of customers that would deploy exchange servers on sands Was enormous. It was the dominant way that you would do it the only way that you could deploy a A classic, you know microsoft cluster, right? Uh, maybe not the easiest thing in the world, but that's how people did it If you look at the last few years the majority of how microsoft has been deployed is basically exchange exchange dag, right so Their own distributed availability groups where they're layering on top of just this inside the server, right? Now no one in their right mind would call that a hyper converged thing Right, but basically the the intelligence is being done higher up inside the application stack another example of that is our obviously customers that Deploy hadoop clusters using hdfs You know as the sds right on top of these, uh, you know simple daz models Now if that lead into why we're seeing some of these bare metal offerings to be able to put some of those solutions into You know a rack space for a public cloud Yeah, and and to be able to manage those low-level bare metal things at higher Higher scale means you have to have scalable management models telemetry falls all that sort of jazz, right? Lots of stuff going on there a place where we're putting a lot of rnd into vx rack and other things, right now This space here You know it's interesting, but then there's there's another thing going on which is hot hot hot and everybody's talked about it Which is basically everything that's going on in the all flash array space right You know the traditional Traditional stores your blue circle is kind of shrinking That market is under compression But there's parts of that market that are booming like the all flash array market is a booming market Um, it is generally something that customers select because they like low latency So it's higher in performance than the hybrids certainly um But the bigger issue the bigger driver is very simple performance So performance that doesn't require Tearing doesn't require caching mechanisms. It's basically a I did a blog post on this where I said it's like the iq test Right, you know if something is faster simpler easier easier and less expensive Why wouldn't you do it? Absolutely what we've seen at wiki bond while the price of flash You know compared to disc has come down when you talk about some of the You know things like compression and dedupe and everything else like that The price gets close enough that Say service providers or certain large companies often will say the operational simplicity of just applying all flash Is enough to make up for any economical difference? And therefore we're going to go all flash whether that's an afa architecture or buying some other box And just putting in all flash and just getting just out of the environment And we'll get to that one in a second Which is kind of like what's going on around the sds space right But there's another there's another interesting one up here, which is a weird little like You know, which is which is pcie and that's ngme for those that can't read at this distance. Yeah, so non-volatile memory Of all sorts of types Lots of interesting stuff going on, you know 2017 will be the First time that intel obtained and and crosspoint, you know start to arrive in the market And for people that haven't looked at it This means that we're getting really close to the compute It's the compute and the memory storage layer That's kind of blurring and rather than having some external box This is something that's real close and we're going to have nvm e f for nvm f I think they call it is so fabric versions of the x Which is kind of new versions of uh, yeah of networking storage Non-volatile memory types, you know exist in all forms So the first example that you point out is basically where you're directly attaching it to the local pcie bus of the host, right? It'll actually show up in two forms one of which will be You know a disc device not a disc device, but a persistence device The other one that will be interesting this year Well, where it'll show up and it'll interface via memory semantics So it'll look like a dim, right? But it'll be a dim that even if you shut off the power the data still stays persistent So it'll take a little while Hence why you're saying persistence rather than storage because our previous definitions of memory and storage are starting to blur They're becoming very blur, right? So um, and then there is an nvm e o f so nvm e over a fabric Uh, which is a standard, uh, that is emergent Uh del uh del technology del emc is playing a big part of that along with intel and with others Um, you know, we think that it's going to be it's interesting. Will the fabric emerge? We're not quite sure Just you know, again anyone who thinks that you know your vendor ecosystem your partners have got all of this stuff all figured out That's not how it works if you invest in multiple areas and you see how things go so obviously When you've got a non-volatile memory attaching it with so low latency and so high performance Attaching it directly to the host bring them directly adjacent to compute makes a lot of sense But then at the same time, it's a bit of an esoteric high-end expensive thing and pooling and aggregating Things that are expensive Also makes sense. Yeah, the only thing I'd say at least we have in some of the early deployments Functionality might be a little bit less than some of these boxes But right it's super high performance kind of a niche inside this market and the smaller As we would say kind of that that tip of the pyramid where some of these really really big companies are doing it I just said intel yourselves and others are doing it. I know wikibon cto david floyer When he's looked at the options out there that nvme nvme over fabric, he's quite bullish on it Yeah, and in terms of the feature set the all flash arrays, whether it's rich remote replication Whether it's deduplication compression Unbelievable snapshot capabilities at this point For the most part they match the traditional storage, you know circle in this weird Venn diagram You know covering the the functionality but always in this higher end of the performance path Now the part of this world that I find personally the most interesting Is I wish I had a different color. I mean, let me grab one blue. You know, I'm gonna grab a green So there's a circle which is surprisingly broad I know this is a little bit different than your guy's point of view of this circle Well, if you're looking at a diagram we drew three years ago This is the the world of software defined storage models and software defined storage models comes in multiple different Ways of consuming it, right? There's just getting it as software So here's the bits Deploy them on whatever you want including things that you send me Inside hosts and it turns it into some sort of form of shared storage, right? There's another one which are server sands server Sand or hci and and then the third version is really this hci version I think that there's a difference between these two and there are and that's why when we created the server sand We're kind of blurring the lines between these what the the big piece is a software only sds could be Only storage. Yes server sand. We've tried to say really it's compute and storage together, which things like up here That could fall under didn't flare didn't want to call it flashes memory extension versus hci A lot of people thought oh, it's just an appliance as a box and we were trying to go beyond yeah so the thing that's fascinating is is the This market is much smaller than the all flash market So if the traditional storage market is what's called seven billion dollars billion dollars You know the all flash array market these days would probably i'm saying this off My highlight say it's probably what a seven billion dollar market sounds about right rapidly growing So it's cannibalizing a lot of the hybrids that are out there This sds market in all three of these forms is still much smaller than that But remember my first comment, which is that our belief is that today sds models regardless of which one of these forms of packaging that you use it in Can support the majority of workloads today, whether it's in this performance dimension Or whether it's inside this functionality then and when you say much smaller, it's about half the size It's not like an attempt of the size of it converged infrastructure is bigger than a phase. Yep That hci everything else like that. It's in the few billion dollars, but it's not, you know Five to ten so so the thing that's fascinating is the reason that customers go this software defined storage route At least in my experience the biggest single reason is It's simpler and easier for them to deploy scale Life cycle manage migrate And not a little bit easier a lot right The second reason is is that fundamentally that the scaling model is so simple So again, whether it's software only or service and hci model where you're getting the components together You know literally the the process of Expansion is not I'm going to add something new. I'm going to need to figure out how to integrate it it's Attachable. Yeah, I grew up. It's the the line. I heard once is the first time you deploy this is the last time You need to do a migration because as anybody that's been in the storage industry knows It's not migration time lost effort is so difficult Now people then ask well chat if you're saying this so emphatically Uh, then does that mean that there's no room for ci or traditional array architectures? And the answer is Well first things first even if today right today All workloads could run on it. It would take time for customers and industry to shift and to move We think that we have a responsibility actually to push The industry to to go and say you can move forward with confidence. This is no longer You know something weird or esoteric. This is something that we as a one of the leaders Uh in this space we're saying it's time to it's time to move now There's a subset of this blue circle of functionality that the sds don't have So people are like well, what do you mean? I'm like well, if you came to me and you said I have a workload That needs to have 10,000 synchronous replicas um, and you need to be able to Do a failover or have a three-site failover behavior and be able to do uh A failover and a fraction of a second um My common would be One you should try to build that sort of availability in your application not your infrastructure And then the customer would go. Yeah, I don't have that choice because this is an application That's been developed over the last 10 15 years. There's no Way I can re architect the app That's the domain of basically all flash arrays or the top part of this traditional storage Were things like srdf and they're very very mature replication technologies Rule of the day Yeah, and talk a little bit more about from the application side You know most applications that I could run in traditional storage can run on the new ones. What about new new applications? How are how are these options? You know here for those those new applications, you know What you see basically is that for any new application it gears towards this cloud storage or hdfs type storage models this Or this right Because you want something that's as simple as possible and you know if anybody is running a new application They're building resilience into the application layer of the stack as opposed to inside the Infrastructure part of the stack again. It may sound like heresy coming from somebody who you know Has been a long time proud emcee here and now del emcee employee But the reality of it is is that we're not hung up on it. We have the strongest software to find storage stacks in vsan and scale io We have uh The strongest cloud storage stacks ecs is really an sds. You can see they kind of overlap We've got the strongest portfolio for customers that are going to deploy things on das using power edge So del emcee power edge We have the strongest all flash arrays and vmax all flash extreme io getting This doesn't intend to be a commercial right But the point is is that the reason that we have this variability is is because customers Are all a little bit different. Yeah, and I mean you throw in here in 2017 You're going to have the address that solution that I know you're going to support from the del family We've got vmware on aws fitting into this mix. So um The picture seems to only get a little bit more complicated. So I guess as we wrap chat, you know What's what's the kind of you know high-level advice? You know you and your team giving to cio's how do they help make this decision other than you know come buy it from you well, so That's actually a good idea So in all in all seriousness what I would say is to any cio who's worried about this stuff They've they've they've got to look in a slightly different place. We live in an interesting geopolitical world right now They've got to be thinking about their own businesses The reality of it is that there's tumultuous change that's occurring inside the industry I think that uh, frankly my my Request to any viewer to anyone that I would interact with is start to become pretty binary about what You do that uniquely innovates for you and what stuff you should frankly just consume Whether it's on-premises for off-premises if you want to deploy vmware on-prem on hci We could deploy that in a simple and easy turn key way if you want to deploy an aws We do that too. You want to have an on-prem azure stack great We've got your back You want to deploy it in top inside azure and you want to put cloud foundry on top of that to simplify further We've got your back The the point is that this rate of innovation is something that we need to drive So that frankly's customers shouldn't really care Right and uh, you know, that's a primary focus for us make things simple All right Well chad really appreciate you taking off the jacket pulling out the marker sharing with our audience You know the rate of change is going on and lots more we expect to see through 2017 So thank you so much and look forward to seeing you at many industry events throughout the year Thanks for watching. Thanks everybody