(techy music) >> Hi, everybody, welcome to this special Cube conversation, my name is Dave Vellante. I'm very excited to be here in our Palo Alto studios in the heart of Silicon Valley, the innovation hub of technology. In 2015 we introduced a company to your community called Datrium and one of the co-founders, Brian Biles at the time, came on as one of our segments and shared with us a little bit about what they were doing. Well, several years on, three years on, this company Datrium is exploding and we're really excited to have Brian Biles back, who's the co-founder and chief product officer at Datrium and he's joined by Sazzala Reddy, who's the CTO and another co-founder. One of the, two of the five co-founders here, so gentlemen, great to see you again, thanks for coming on. >> Good to see you, Dave. >> Yeah, so Brian, I remember that interview and I remember, you know, trying to get out of you what that secret sauce was, exactly what you were doing. There were a lot of other start ups, you know, at that time and several have gone by the wayside. You guys are exploding, so I want to help people understand why you're being so successful. Now, I want to start with the two co-founders. Why did you and your other co-founders start the company? >> You know, we started the company... We hired our first people in 2013, and at that time, there were really two separate worlds. There was a cloud world and there was an on-prem world that was sort of dominated by VMware. So, there were these two evolving discussions about how each one was going to grow in it's own way but kind of within its sphere. We thought there was an opportunity to bridge the two, and to do that, you know, ultimately it becomes a questions of how to run sort of coordinating applications on public clouds and deal with data on public clouds but also to have a capable version of the same infrastructure on private clouds. So, our sort of job one was to build up, to sort of import cloud technology onto prem. We currently have, if you want an Amazon-like version of infrastructure on-prem, we're still the best place to go because we have a two layer model, where there's, you know, compute with fast flash, talking to a separate durability layer very much like EC to an S3. You want to do that, we're still the way to go. But the long term story is also developing. We have a footprint on cloud with a backup store on S3 that coordinates all our data services for global deduping security and so on in a very cost effective, simple SAS way, and that part is growing significantly over the next couple of years. So we're, you know, through with sort of phase one. It'll keep, you know, evolving but phase two is really just getting going. >> So Sazzala, as the chief technologist you had to think about the architecture of where the industry was going and the architecture that would fit that. And you know, that people talk about future proofing so if you think back to the original sort of founding premise, what were some of the challenges that you were trying to solve? >> Right, so there's a business use cases and then there's technology use cases. And as a CTO you have to think of both of them, not just technologies, so if you look at technology point of view, you know, in 2000, back in 2000, Google published a paper called Map Reduce that said hey, this is all we can do at large scale. It was the beginning of how to build large scale distributed systems. But it was built for one use case for surge. But if you look at, we started in a time when Google was already there and they built a system for multiple, unpredictable use cases. So you think differently how the problem is whereas Google start from, though. Some of the CI vendors, they've done good things. They kind of evolved in that direction. We have evolved in a new direction. To the technology point of view, that's kind of what we thought about. But from a business perspective, what do people want? You know, if you look at the next generation, the millennials, and look beyond that they're used to iPhone experience. They don't want, if you tell them about LUNs, they don't re-phone LUNs, they're going to just say what is this and why do you have this stuff, right? So you have to evolve away from that. So, it's the CIA wants to think about how do I make my idea the service? How do I consume, you know, how do I make it a consumption model, how do I make my IT not a cost center but a friendly way to you know, grow my business? And the developers want a platform they can develop things faster, they can adapt to newer, kind of newer technologies coming in, there's Mesos, there's Docker container, there's Kubernetes, thus things change rapidly. So that's going to build a framework in how we wanted to start the company. Basically build a cloud-like experience simple as a SAS, simple as a click and then just make that work. >> The thing that's interesting to me about Datrium is you know, the simplicity, like open. You know, I remember when Unix was considered open and then obviously the definition changes, simplicity has changed. I remember when converged infrastructure, bolting together, compute storage and networking, simplified things. Hyperconverge took that to another level. You guys are going beyond that taking it to yet another level of simplicity, so I wonder if you could talk about that-- >> Yeah, so-- Specifically in terms of the problems that you're solving today for customers. >> So if you look at the V block, I guess the VCE was the first, I guess that they made a successful convergence. So they did hardware convergence-- >> Right. which is a useful thing to do. Same thing with your head CI, the traditional vendors, they do hardware convergence, if you look at Hecht, probably stands for hardware convergence, maybe. But we took a little bigger step in the sense that what you really want us to think about is data convergence. Not, hyperconvergence is useful, but you also think operating about data convergence. What's the point of building your on-prem cloud- like experience when you still have to do backups and some other you know, some other boxes you are to buy. That's not a really good experience, but you need is this whole new hardware convergence, we also need data convergence to get that experience of like cloud-like simplicity in your on-prem. >> Right, in the cloud you don't think of backing it up, right, it's self protecting. That's just the nature of how you should be thinking about on-prem as well. So, when we imported that technology to be a two-layer approach, we built that stuff in so you don't have to think about it. It's kind of like there's no SQL or we're sort of like no backup. >> Yeah, we're going to talk some more about that but that's an important point is you get backup and data protection, you know, full capability, it's just there. I always use the example of Netflix or Spotify. I don't have to call up a salesperson or the billing department or the customer service department, it's just there and I deal with it. >> Right and it gives you, you know, this combination of, in the two layers, the ability to run multiple workloads at big scale, which is otherwise hard in some of these more historical approaches, with great performance that you know is off the charts. But it also means you don't have to move data around as much. So you restore, you restart, you don't restore. You don't copy stuff in and out. >> Yeah. >> That data mobility efficiency it turns out, is also super critical when you think about multi-cloud behavior. >> You have to be in the business to actually feel like you talk to backup admins and life is hell. It is really painful and it's also very fearful if you have a problem, you have to restore and everybody's watching you when you're restoring. So we try to eliminate all those problems, right? Make it, just, why worry about all these things? We are living in a new world, let's adapt to it. >> I think I've, tongue-in-cheek I think about the show Silicon Valley and you guys didn't start out to build a box. >> No. >> No. >> You settled this off some problems and so what you have is a set of best of breed storage services that are running the cloud, called multi-cloud, meaning on-prem or in the cloud so I want to try to juxtapose that to sort of the traditional storage model or even some of these emerging storage models of some of the very successful companies. So, how do you guys differentiate, help us understand what's different about Datrium from the classical storage model and even some of these emerging storage models? >> I'll kick it off and Sazzala can expand on it. You know, first we're bringing a cloud experience to on-prem, so it's not a storage system that you, like a SAN. We, you know, offer compute as well and a way to make that whole operation simple around you know, standard and emerging standard coordination frameworks like VMware and Red Hat and Docker. It includes these really powerful data services to make life simple so you don't have to add on a lot of different control panes and spots of data storage and so on. By getting that right, it makes multi-cloud coordination a lot easier because the hardest problem getting started in that, aside from, you know, just doing SAS applications to run it and so on, is getting data back and forth. Making it efficient and cost effective to move it. So, you want to expand? >> Yeah, so you know, I think you give examples of like maybe there are some successful companies in the market today. There is old school array market and there's the new school head CI markets. So, the old school array market, I mean, if some people are still comfortable with that model, I think they just because the flash array market has some performance characteristics but still it's, again, going back to that rotary phone landing, it doesn't map your, the lands don't map your business. It's just a very old school way of thinking about it. Those will probably vanish at some point because it makes no sense to have them around. And yes, they do provide higher performance but they're still, you know, it's still not providing you that level of ideal service. From a developer point of view, I can make my application life easier, I can do things like test and dev. Test and dev, simple thing like test and dev requires you to clone your application so they can run test and dev on them. It's a very powerful use case, it's a very common use case for most companies, including ours. So, you can't do any of that stuff with that old school style of array. And the new school style, they are making progress in terms of making that developer life a little bit more easier but they haven't thought deeply about data services. Like they built a nice packaging and like some UI frameworks but ultimately, data needs to be like stable. They didn't, you think about data in a how do you make it compressed, efficient and cost effective and make it so that it is easy to move data around. And you're think about the backup and DR. Because if we look at application, you've run it, you have to back it up and you have to do archiving for it. You have to think up the entire lifecycle of it. Which is kind of what most people are not doing, thinking of the entire lifecycle. They're solving a small piece of the puzzle but not the entire thing. >> I'll give you another example of that. In you know, to the operator of a private cloud, you're thinking about workloads, you're thinking about relationships between VMs you know, how to get them to the right place, copy them at the right rate, secure them in the right way. In a sort of old style, that kind of thinking about say protection, you might have a catalog in a backup software but you have volumes of VMs in a SAN. Those are completely different mindsets, we've merged them. So we have a completely scalable catalog you know and detailed validation, verification information about every scrap of data on the system that we can test everything four times a day for test restores. All that kind of stuff is organically in a single user interface that's VM focused, so you don't have to think about these different mindsets. >> But it's SAS really for data services. >> For data services, yeah. >> I mean is that a fair way to think about this? >> Yeah, I think so because what's better than one click? Zero clicks, so lot of people are aiming for one click. We are aiming for zero clicks. That's actually a harder problem to do. It's actually hard to actually think about how do I automate everything so they have to do nothing? That's kind of where we have really, really tried hard is that, as little clicks as possible. Aim for zero as much as possible. That's our goal, in the internal company engineers are told you must aim for zero clicks, actually a harder problem. >> Right so, when you think about how to then expand that to managing multiple sort of availability zones across multiple clouds there are additional problems. But starting from these capabilities, starting from great indexing of data, great cataloging of relationships between things, everything's workload specific and great data mobility infrastructure with data reduction and encryption and so on. As we forecast where we can go with that, it's profound. You can start to imagine some context for how to deal with information across clouds and how to both run and protect it in a way that's really just never been in the market. >> So I want to talk about that vision but before we do, before we leave sort of the differences let's take two examples. Two very successful companies, Nutanix and Pure, so how are you different from, let's start with Nutanix, for example. >> I think that there's some good things, I think they're moved the industry forward quite a bit. I think they've brought some new ideas to the market, they made it VM-centric, they said no LUNs. They've made quite some improvements, and then they're a successful company, but ultimately I think their focus tends to be mostly on how to make the UI shiny and how to kind of think about the hypervisor, which is kind of where they're going to. They don't hypervise in the world today, we don't want to go invent another hypervisor. >> Mm-hmm. >> There are so many other options and the world is changing a lot. Like you said, Kubernetes is coming, Mesos is coming, so we want to adapt to those newer ways or style of doing it, and we don't want to invest in making or building a new hypervisor, and we're good partners with VMware, so that's one angle to it. If you look at, you know, how... Because if you're going to go to large enterprises, they want to consolidate the workloads. They want large scale, they want exabyte scale, so you meet customers now who have exabyte scale data, they think they're the cloud. They're not thinking of any other cloud, they think they're the cloud, so how do you make them successful? So, you have to think about exabyte scale systems where basically you can operate it as a cloud internally, so we build those kinds of infrastructures and those kinds of tools to make that exabyte scale successful, and we probably are the fastest system on the planet. Right, so that's kind of where we come from is that we not only say that we scale, we actually prove that we scale. It's not just enough to say we have Google style and the scale, so it's actually you have to prove it, so we actually have tests where we can, we actually have run with other people that it actually works as we say it does. So, I think it's important that you have to speak, you have to not only produce a product which is useful from a UI point of view, that's useful, but also it has to actually work at scale, and we make it more resilient. We have a lot of features built in to make it more resilient and at scale, like what does a tier one mean, what is mission critical apps, how do you make sure that we don't lose data, for example. It runs at the highest performance possible at a price which is reasonable. >> Okay, and I guess the other difference is you're a pure SAS model in that you're responsible for (chuckles) the data services, right, and-- >> Yeah, that's right. >> Yeah, we've pulled a lot more into the data services in our cloud approach. >> Mm-hmm. >> And we've separated from the performance elements, so they're these two layers, so it's both self-protecting in a way that's independently provisioned if you want to expand capacity for backup retention, that's a standard thing. If you want to expand performance or workloads you do that independently on stateless hosts. >> Mm-hmm. >> An example of where this pays off is just the resilience of the system. In a standard hyperconverged model a good case is like what's the crater size when a, or the risk, you know, profile when a single component fails. So, if a motherboard fails in a sort of hyperconverged model that's standard, you know, a single layer thing, then all the data on that system has to be rebuilt. That puts enormous pressure on the network, and you know, some of these systems can have 80, 160 terabytes of data on a single node, that's like a crazy week, and if two of them go down then the whole thing stops. In our model the hosts are stateless, if any number of them go down for any reason the data's still safe, separate-- >> Mm-hmm, right. >> You know, in a hyperconverged model you can't really integrate backup well because when primary goes down back up goes down, too, then what? >> Okay, so that's, I think, clear how you differentiate from hyperconverged. Did you have another-- >> Yeah, I have one more point, it's about the data services you mentioned. We have, again, going back to zero-click, we built all our features into the system. For example, you know, there are a lot things like deduplication, compression, image recording, those are like, I mean, they're not like details, but ultimately they do bring the cost down quite a bit, like by 10 X to five X, right, that's a big difference. >> So, those are services that are inherent. >> That are inherent in the system. >> Yeah, okay. >> Either you can have check boxes, you can say one click and have to like check box, all that. I mean, you have to go and click it, but to click it then you must read a manual, you must do the manual, so then what is this right click and what happens to me, why isn't not on by default. >> Yeah. >> So, those are the problems, I think the differences between them, I think Nutanix and us, is that we kind of made it all, like, be seamless and all built in. >> Yeah, and when we, you know, if you have to, if it's an option that you ask for later that means it probably has some impact on the system that you have to decide about. In our case you can't turn it off, it's always there and we do all our benchmarking with all that stuff turned on, including software-based encryption. It's just a standard thing, and we still are like the fastest thing on the planet. >> Yeah. >> And let's talk about Pure a little bit, because they don't have-- >> Yeah. >> The networking component and then the compute component, it's, you know, flash array, so how would you position relative to Pure? >> Okay, so again, going back to that SAN array was built before the internet, it is just the same. It is just the same, it's just to deport SSDs behind those controllers in central hard drives. It is likely faster, but ultimately the bottleneck is those controllers, those two controllers they have, that's what it is. No matter how many, how awesome your... You put envy in drive, it doesn't matter. It's going to be as much as speed as your network pipe is going to be, and as much faster as your controllers are going to be. Ultimately, the latency, you cannot, like, basically it's over the wire. It will always be slower than what kind of having... >> So, the big thing here is-- >> Yeah, and it's not a private cloud. You know, that kind of model is for someone who's assembling a lot of parts to create a cloud. >> Yep. >> You know, we're integrating these parts, so it's a much simpler deployment of a cloud experience and you're not integrating all these double parts. >> I'm getting a cloud, I'm buying a cloud experience from you guys with the sets of services, let's talk about those services. So, mobility, discovery, analytics. >> Yeah. >> Governance, talked about the... >> Encryption, yeah. >> The other data reduction services, encryption... >> Right, the cataloging and indexing of the data so you can, you know, restart from old data. >> And I can run this on any cloud, including my on-prem cloud, correct? >> Well, that's the direction, we have some parts now and you know, you... (laughs) Sorry, Sazzala can talk about where we're going. >> So, architecturally it's designed to run on... >> Yeah, because I think fundamentally we chose that design philosophy that it has to be two-layer, right, that's a fundamental decision we made long ago, and it's a detail but it's a fundamental decision we made long ago that because if you go to Amazon it is two-layer. You cannot make one-layer work there. Like, you know, compute and storage has to be split to through that part, but they must work together in a nice way, and also S3's very weird. I don't know if you know about S3. S3's very weird behavior, it does not like random writes, it has to be all sequential writes, and that also happens to be how we built it. The way our system works is that we only do sequential writes to any device. It works beautifully in S3 with EC2, so just to step back a little bit, taking big picture, like so, we wanted a cloud-like experience for your on-prem, right. That's kind of what we built, we built a Datrium cloud on-prem, and then we, as of beginning of this year, we started offering services, multi-cloud services and started with Amazon first. The first service we enabled was backup and archiving, that's our first service. A lot of people like it and you have some stats from that, like from last quarter, like how people like it, because people like it because you don't have to have another on-prem infrastructure. You can just consume it as a SAS model, it's very convenient and it's as easy as an iPhone backup. I don't know if you use iPhone backup, it's like a click. >> Yeah. >> Okay, unfortunately it's a click. We have tried to avoid the clicks, but we can't really avoid it all the way, so you have to click it so that you can then start doing backups into the cloud and then can retrieve them in a very simple single pane of glass. It's very cost-effective because we do dedupe on the cloud and we dedupe over the wire, but dedupe over the wire, by the way, it's actually a very unique feature. Not many companies have it, like Nutanix and Pure you mentioned, they don't have it, so you know, so that's one of the things where I think we differentiate because data has gravity, right, so to move it somewhere you need an antigravity device. So, you need something to actually move this data faster, how to defeat speed of light. You have a pipe, you have a VAN network, so how do you defeat the speed of light, so what we have built is a feature, it's called Global Dedupe, is that you can move data in a much more efficient way across the cloud. So, now you may question, "Hey, I'm moving my data "from here to another place," obviously we have these cloud services... The question you may ask is, "Okay, how do "I know I get guaranteed security? "How do I know that it's going to be correct, "that I moved all these places," right? So, we do multiple things, one is that we have built in encryption. It's going to be globally encrypted, it's like an encryption across the whole thing, we call it blanket encryption. >> Mm-hmm. >> The other one is that we have blockchain-like features that are built into the systems so that if you move an object, like an app or whatever, you're going to move from one place to the other, it's built in kind of blockchain features where you cannot move something to another place and get it wrong. It's fundamentally going to be correct for you, so those are the kind of things we thought about, like never to worry about it again. It's going to guarantee the data's correct and it's moved in the most efficient way, so that's our first landing thing we've done is that we wanted to build an experience which is like on-prem cloud, I mean, onto also the cloud. Right, what other experience people are... People like simplicity, people want the SAS-like experience. They don't want to manage it, they don't want to think about it. They just consume the services, so the first service we have in Amazon is what we chose, is backup and DR. The next thing we are going to be shipping soon, announcing soon, and we'll have a demo in the VM World is something we call Cloud Shift. It's an app mobility orchestration framework where you can just click and move your workload to somewhere else, to Amazon, and you can run, so it's not just a backup thing, it'll also become you can run your workloads in Amazon and get a consistent experience from your on-prem and the cloud. So, one of the challenges is that if you move to another place, is it different tool sets, I have to change my whole lifestyle, no. >> Mm-hmm. >> We want to provide that seamless operational consistency that-- >> That's the key, right. >> That's the key. >> Whether it's on-prem or it's in the cloud it operates the same way. I'm accessing those sets of data services and-- >> Yeah. >> I don't really care where it is, is that-- >> That's right. >> The vision? >> Yeah, that's right. >> Exactly. >> That's right, so if it turns out that there's a cost advantage in moving from, you know, A to B, we make it super easy and the control panel from our standpoint is consistent, and it's... So, all of our control orientation moving forward will literally be SAS. It'll be running on a cloud even if you're managing on-prem stuff, because that way, assuming you're multi-cloud, you need a control plane to be dealing with the cloud stuff anyway, and it just sort of neutralizes the experience so that in a multi-cloud way it's always consistent, it's always simple, and the nice thing about sort of true SAS is you don't have to upgrade software parts. We do that for you in the background. >> Mm-hmm. >> So, it's just always up to date. >> So, I was saying before, Datrium takes care of everything. >> Yeah. >> And it's the true cloud experience. >> Just consume it. >> Right. >> Okay, I want to talk about, end on the two other areas: the operational impact and the developer impact. So, when you think of operations, we've talked about LUNs before. I've always said if you're in the business of managing LUNs you really want to think about, you know, updating your skill sets (chuckles) because that capability is not really going to be viewed as valuable. It isn't today and certainly in the future, so the operational impact, the degrees of automation that IT operations are driving is going through the roof. Cloud-like, we've talked about that, and the other is developer productivity. People are using containers, you know, Kubernetes... >> Yeah. >> And new styles of writing software-- >> Yeah. >> As everybody becomes a software company. So, can you talk about those two aspects? >> And ultimately there's going to be serverless. >> Right. >> Right. >> As we think about if you take a leap, in another 10 years I think serverless will probably be one of the important ways, because why do you even care how it runs. You just write some software and like, you know, we can run it. It should be that way, but I think we're not there completely yet, I think, so we want to adopt a methodology where we provide the framework where we don't dictate what apps, how we write your apps. That's, I think, very powerful because that's actually evolving faster as we move forward, because serverless is a new app framework. >> Mm-hmm. >> You cannot anticipate this, right, you cannot anticipate on building everything but what you can anticipate is services we can provide for the developers, which is, you know, no matter... Because it's the granularity of it. We can map their application granularity into our system, we have that fine level granularity, so that kind of was what you want to provide as a primitive. LUNs don't have that primitive, right, so we provide that level of primitive that whatever apps you have will have that level of primitives to global data services for you, and once you have the data services like that we'll guarantee that it's highest performance, which is what app developers want. Like, I get the highest performance, I can easily... And then we will also provide a way to clone those things easily, those apps, because sometimes you're at an app, you want to test it, too. Like a hundred times, you want to just... If you can copy all the data a hundred times or you can just, say, you know what, clone this thing a hundred times in a millisecond and run my tests fast and then okay, I'm done with my test, it looks good, I'll deploy it. >> Mm-hmm. >> That's kind of what developers really want is that they are able to run, write faster, develop faster, because tests on dev cycles are important. A lot of people think that hey, I can put my test on dev in some old box over there, but that's really bad because from business perspective testing does, engineering's expensive. Their test cycles have to be fast so that they can e-trade faster and kind of produce faster. The harder you make it to test your system, this is like, this is what happens in our company today. The harder it is to test your logic and your code, the longer it takes to, like, do e-trade. >> In some ways test and dev is becoming more strategic than the production system, I mean, really-- >> Well, it-- >> (chuckles) Because of speed. >> Yeah, I mean, it can take immediate advantage of some of these improvements in, you know, stacks. Like if, you know, if Kubernetes is better just, you know, go quickly to it. The things that these new stacks assume, though, is that it's, you know, a server-based data, so on-site you can accelerate mobility significantly by, you know, when people ask to copy things from here to there, clone it, you know, start another instance, we can help them do that by just, you know, faking it out with metadata-- >> Mm-hmm. >> And deduplication, and so we tried this with Jenkins just in our own development, moved to that model and you know, everything was suddenly twice as fast in development. To do a build all of a sudden you didn't have to copy data here to there. You were cloning, you know, with metadata. The way to do it across clouds is, again, kind of dedupe focused. If you have to actually move the data it takes a long time and it's expensive, especially for egress costs. If you can just, you know, validate which elements of the data are new versus old on either site you can move a lot less. >> Hmm... >> It might be, you know, six times less, and then the costs go down, the speed goes up, you defeat data gravity. >> Yeah, so-- >> Excellent, all right, we have to leave it there. >> Okay. >> Out of time, thanks so much, you guys, for helping us better understand, you know, Datrium. Congratulations on your success so far and all the great innovations that you've achieved. >> Okay, thank you. >> Okay, thanks for watching, everybody, this special CUBE conversation. This is Dave Vellante, see you next time. (techy music)