 Live from New York, it's theCUBE. Covering Micron Summit 2017, brought to you by Micron. Welcome back to the Big Apple, everybody. This is the Micron Summit, and this is theCUBE, the worldwide leader in live tech coverage. My name is Dave Vellante, and I'm here with my co-host, David Floyer. The Micron Summit's been a day-long event announcing a number of new products. We started off with Darren Thomas, was hosting Laura Dubois from IDC, gave a big picture overview of what's going on the market. We had CIOs and practitioner discussions, but really, right now, we're gonna get into the product, into the NVME and NVME over fabric, and what was announced today, a new product called SolidSkill. Eric Enderbrock is here. He's the Vice President of Storage Marketing at Micron, he's joined by James Meeker, who's the Director of Enterprise Solutions. Gentlemen, thanks for coming on theCUBE. Well, thanks for having us, I appreciate it. You're welcome. So, Eric, let's start with you. I was picking up on what you said. You had talked about, hey, if you're gonna be at OpenStack next week, which by the way we are, theCUBE will be. I was yesterday at Red Hat Summit, and it was a big discussion about workloads, and of course, OpenStack's a big component of Red Hat's business, and they were talking about modernizing workloads, how workloads are diversifying. Many, many examples, whether it was a web sphere, a web logic, they even gave a JBoss, you think JBoss is pretty modern, but no, things are changing so fast, so maybe start there. What's changing in the world of workloads, and how is Flash supporting those changes? Yeah, and in some cases it might be what is. We, for a lot of very valid reasons, we made really fairly smart and intelligent memory look like a hard drive, a spinning media, and we did that so we didn't have to rewrite code, and a lot of great reasons, it makes substitution great, but it leaves huge amounts of performance capability just left orphaned on the table. And we do some work in Austin, a lot of software development, and one of our key pieces is actually how do we break through that for workloads, and what we find, we call it write amplification, but every time I wanna write one byte onto an SSD, by the time I get up to an application level, to all the different things, file systems, the application itself, even the firmware on the SSD, you'll find that I'm probably writing hundreds if not thousands of bytes of data all the way down at the very bottom of that stack, and so that's a piece where we've got to, as an industry, break free and unleash some of the value of flash, because that's the next frontier. If we can create a much more direct access layer from those workloads, they're gonna get a lot more performance out of it. And that's just one piece, a long answer, I guess. No, that's great. But you think about what that looks like, and that was a storage question you asked, but it's the same question if you think about memory, and in memory databases, and where these applications and workloads are going, we've gotta fit more in memory, which is gonna take memory extension technologies, many of which are based on flash and 3D crosspoint and other technologies. And then at the same time, I've gotta have the storage that's fast enough and big enough to feed them, especially as these algorithms start creating that data. So James, you're in Austin. That's right. A lot of the software is being developed. So what's the crux of the initiative there? Is it to eliminate what I referred to this morning is the horrible storage stack? Well, we've been focusing on how do you take a great technology like NVMe over F that has super low latency, and how do you get that performance, that capacity outside of a single server and share it with all of your servers and all of your applications. Latency typically will get lost across your fabric and applications won't be able to take advantage of it. And so one of the big things we did today was introduce solid scale. And solid scale is an architecture that we're delivering that allows you to share NVMe low latency benefits, high capacity across all of your servers. And we focused on that is the key product delivery today. So how would I, as an end user, take your product and actually utilize it? What are the steps that I would go through to actually to get some value out of it? Sure, sure. Try it out. So for many customers they'll have a specific application or a workload profile that they'll want to deploy this technology on. In many cases you've got workloads like analytics, big data type applications that you'd be able, you would want to architect around specific densities and performance targets. The solid scale structure is flexible enough in not only how you deploy it, but how you architect it, that you can scale your storage in a node-to-node scale out type configuration or have a consolidated centralized storage configuration for flexibility with your application deployment, depending on really what are the needs of your users and the IO patterns that you're trying to improve. Could I jump in real quick? So a piece of that though is, I think James did a great job describing the architectural view of that. Very mechanically, we see the easiest way. Most customers are deploying rack scale these days. So in these early phases, we're producing this on a 24U rack. And so we'll roll it in. It's got top of rack switching. So that connect out to your general purpose fabric. And you've got to be very flexible with that. People have fabric choices they want to make. And then you have your Melanox infrastructure within that, the solid-scale appliance piece itself. And then you still have quite a bit of room course power and PDUs and things like that. But in the middle, then you've got space to put in compute if you choose to do it there. And again, a very customer-centric choice, not trying to necessarily define what their compute architecture looks like. So I think we've made it very easy to get this into an environment physically, very practically. Yeah, exactly. Plug it in, get that part running. And then if you choose to, very easily connected out into your broader network, but the key is keeping that very high-speed, low-latency traffic kind of isolated to where you need it. That's an important part. So one of the areas that when I've talked to people about this is the excitement is being able to combine transactional systems with the analytics, the artificial intelligence, the background. And combine it in two ways. First of all, to be able to share the data between the two so that you have a different view of the data, but this is actually the same physical data that is too logical, looks at it. And then secondly, to be able to in real time get the analytics over to the transactional system itself. So to me, the business value of that, all the people I've talked to, the potential business value of that is enormous in terms of what it can do to the productivity of a company, either like Micron in the area of manufacturing or to any other company at all. How would you put your technology, how would you work with your customers to put this technology into such an environment? I mean, you want me to grab that or? Sure. Go ahead. I'll give it a shot. So I think it's a great question and the vision you describe, I think is the one everybody's after, right? If you think of IoT and any kind of data consumption as this massive dump of data comes in, how do I collapse that to the 2% of it that might be actually valuable, the insights hidden within that haystack, if you will, of data and then potentially discard the rest or do whatever you're gonna do with it. But at the same time, connect to that transaction piece. We have a very small but very specialized consulting group and so we definitely work with our customers. Part of how we got to solid scale in fact is a piece of this customers coming to us asking for exactly what you've described. For many years, we've had a team very focused on working with anybody who wanted to build and deploy flash or memory in any way. Again, you can tune flash, you can customize it, you can reprogram it, do a lot of things but you can also destroy it if you do that incorrectly. So we had a team that's very focused on that. We've been building in ritual and bat team as well to be much more workload aware. And so through this, I guess advisory group team that we focus on this area, that's how we generally approach this with customers. Now, they know their apps best. I don't wanna pretend that, you know, Micron's coming in and advising them on how to architect and build their full solutions. That's not the case, but what we can do and what we have tested and what James and a lot of our engineers are doing every day in Austin and throughout our labs is finding ways to break it and finding what doesn't work. So a lot of our value is making sure you avoid just common pitfalls so that you don't actually sub-optimize it. If you're dealing with 200 microseconds, one screw up turns that into a thousand and anybody can get a thousand microseconds out of the system. So, you know, you gotta stay pretty tight on focus if you're gonna reap the rewards. Should bring up a really good question. IoT applications, the analytic cutting-edge type applications need an infrastructure that has these ultra low latencies in order to process the inbound data. And so we've really focused on that as a company, as an engineering group to try to deliver solutions to enable customers to port over their applications and face new challenges that are high speed, low latency with an infrastructure that can scale out to support their businesses and that's what solid skills have been about. So I got a question for the three of you maybe. So I feel like we've been solving the same storage problem for years. If I go back to sort of late 90s, early 2000s, the storage area network, it was supposed to be the centralized pool of storage, sort of improved utilization, improved performance, certainly, because we're off loading the host. It was an availability component as well. When I read your press release today, I said, it's like back to the future. Yeah, exactly. I'm really going to solve it this time is my question, right? So, and I think the reality is, okay, so we made some progress on things like utilization and obviously performance easily was eaten up by the exponential data growth. But what's your take on that? Maybe James, if you could maybe look back and say, okay, this is the industry solves some problems, but boy, they fell down in terms of meeting the promise. I feel like it's, you know, the old data warehouse. Oh, it's going to save the world and 360 to view your customer and it didn't happen. And so, will we get there this time is really my question. Yeah, I think we're really seeing a tipping point in the industry. We've recognized that there's boundaries on how fast you can actually go, how many cores can you put in the CPU. It's about getting your data closer to your processor. It's about being able to feed all of those cores inside your racks of servers and flash is inherently a technology that allows people to move their data closer to the CPU to process more. And it brings a lot of benefits and it also brings a lot of challenges. And so when we start thinking through, would you design flash or storage solution today with flash technologies, if you had to do it all from scratch, if you started out back when hard drives came out or SSDs entered the market, it kind of had to fit the norms of what a hard drive looked like when we launched SSDs. But if you did it today and you know what you know about flash and the benefits it could bring, you would re-architect the whole solution to take advantage of the power and the low latency and the benefits that you get from a solid state data center. And that's our vision. We're going to get it right this time by starting grounds up, by looking at flash as the medium to solve data center and storage problems. And I think one of our CIOs said it right though, the sooner as we solve this problem, we're probably breaking another. And so, you know, and we know that already, we can saturate any fabric that's in existence. Exactly, we're putting the bottlenecks in there. So now I'm just pushing it to somewhere else until we'll jump ahead and do something clever there. So, you know, it's definitely a rat race, but... But we can pound on the networking guys for a while. You know, I don't mind, I don't have storage guys. We don't mind taking the meeting, that's okay. As long as you keep calling. But, you know, and, you know, James and I spent many years in former lives, you know, building storage systems and changing over from fiber channel. Well, actually, I like to think of it as, you always see those evolution things, you know, and you got the, you know, people standing upright and walking around that evolution thing. I kind of think of it in storage world, you know, we were direct attached and dedicated in the main frame. And then we went open solutions and we put it in direct attach, and it was cheaper. Then we went to sands and it was all good. And then we kind of shrunk back to direct attach again. And that's sort of where we've been with flash. It's like, that's how you take the most advantage of it. And what we're trying to do with solid scales, break that free one more time. So, you know, it does, it cycles, I guess. Just answering in a slightly different way. If you go back to the sand days with disk, you were lucky if you could get 10 milliseconds with this, with flash, you've got it down to around one millisecond, one to two milliseconds, when you looked at the real averages. What we're talking about now is 200 milliseconds to the actual data. And within the previous generation of sand, if you went off that... 200 milliseconds, microsecond, microsecond. Microsecond is where we are. If you, in the sand environment, if you went from one CPU to another to the data on another, that was a long, long, long drive there and back. That was many, many milliseconds to do. Now you're down to just five microseconds. Okay, so let me pick up on something in your press release. As I'd like to clarify it here in the cube, because I was a little bit confused and you had said you got it, but let's get out there. One of the bullets was optimized performance. So the speed of the micron NVMe SSDs coupled with high bandwidth melanox fabric delivers performance at scales by adding an average of five milliseconds of additional latency to an application's data path when compared to a local in-server NVMe. Micron solid-scale architecture is expected to reduce end-to-end latency to under 200 microseconds. Okay, can you explain that to the late person? Yeah, well, I'll grab it. So, you know, in our testing, you get about 195 microseconds. And by the way, I got this wrong yesterday with the Millie and Micro. The good somebody, they told me to say, well, just say micron. I said, well, that helped, actually. So I've got it right today at least so far. It's about 195 microseconds end-to-end on that transaction when it's local. So that's what an NVMe bus today does. Now, obviously there'll be improvements there. Sure. A lot of great stuff happening there that we're all over. Our design goal, as I mentioned in the presentation real quick, was to say, how can I create as the same or, you know, not much worse in the worst case, a centralized pool that you could do it? And so we add, on average, five microseconds over being local. And that was what I was saying in the press release is, you know, for a five microsecond penalty, you can unleash that flash from being direct and dedicated to an individual server and get all the benefits of that shared, accelerated storage. The pooling and the sharing. Better management and everything. The much simpler, lower cost of ownership. Absolutely. The software stack doesn't get in the way of your latency improvements. You're going directly to these huge pools of NVMe and across your fabric, and you're getting access to all your remote data as if it were a local drive. And that's what's phenomenal. The software is not in way of the benefits and the goodness that you get from this technology. Yeah, well, so, I'll go, please. Well, I was just gonna say, you might even mention, like, bypassing the CPU in some cases to get the most out of this. So that's another thing that really is a challenge for a lot of customers is that to manage the I-O consumes CPU cycles. In many cases, it can be 20 to 22% of your CPU cycles will be dedicated to managing the I-O on your servers. By bypassing this or offloading your I-O requests, it frees up your CPU, so you get a better TCO, you've got more virtualized machines or more processes running on your clients. And that also not only solves an I-O problem, but it gives a better value to the customer to increase their density of servers in Iraq. And that's game-changing. That comes back to workloads. Eric, you talked about VMware and VSAN at the top, which is now, VMware's the new legacy. Sorry, Pat Gelsinger, but it's true. And we're onto containers and other things, yeah. Right, we're onto containers and Kubernetes and mesospheres and all that cool stuff. And that's a whole new world and you're talking about orders of magnitudes, performance increase and the ability to handle more data, share data. Good stuff. Dave, one thing, you mentioned the workload and kind of circling it back around to that. I mean, one of the interesting things, and I think the phenomenon that may be the game changer that does change this and maybe solves it a little differently than we have in the past is, you take a fortune, any Fortune 1000 company. We're running tens of thousands of applications. A lot of what we're solving for are companies who are running 10 applications. They're more focused apps. They're bigger, they're scaled out in all of those cases that I'm kind of describing, but they're also tunable. And those are the companies that came to Micron and said, you know what? I've got a very unique problem, but I can actually very uniquely solve it because I wrote the code in many cases or I have 100% understanding of how every piece of it works. And so they were really the catalyst for solid scale. And you can call them hyperscales or cloud vendors, whatever, any number of these different kind of companies look and feel like, but their ability to so granularly control their environment is allowing us to do really great things with Flash with them. And part of solid scale was saying, you know what? How do we take that out further? How do we push that to the edge and get it more broadly available and still be unique in value add? You know what I mean? Solving a VMware vSAN, you mentioned its legacy. That was great learning for us. I think we made a great system. It had a lot of value and customers really liked that, but it's a whole new level with solid scale. That takes it to the next level. Yeah, and those are the apps that are changing the world. And like it or not, folks, it's coming. Gentlemen, thanks very much for coming on. Appreciate it. Appreciate it, thanks. All right, keep it right there, everybody. We'll be back with our next guest right up to this. We're live from Micron Summit in New York City. This is theCUBE.