 From New York, it's the Cube. Covering Micron Summit 2017, brought to you by Micron. Micron Summit, we're here at 34th floor of One World Trade Center. My name is Dave Vellante, and I'm joined by David Floyer and Nick Allen of Wikibon. And we've been here talking all day about Micron moving up the stack, innovating, bringing out NVMe, NVMe over fabric. Micron is a company with real tailwinds right now. D-RAM prices are up. Supply is down. There's been consolidation in the industry. Samsung is not bombing prices like they were a few years ago. And as a result, Micron stock is up. They got the existing CEO, Mark Durkin, leaving. New CEO coming in right at the peak time. A lot of good things going on for Micron. The questions are, as they move up the stack, Darren Thomas mentioned this a little bit. We're not trying to compete with our customers. Okay, so you know it's come up. It's certainly some of the folks that I've talked to in their ecosystem have said, what's going on? But as Mark Durkin said, it's inevitable. And I think we've believed that, David Floyer, for some time now. Your take on today, what you heard, what's real, what's to be determined. Well, my overall take is that this is the start of an even bigger disruption than we've had over the last 10 years. So this is, to me, one of the biggest disruptions that is taking place. What's the this, what do you mean? And that is the ability to bring in the fabric side of things so that you have a balance between memory, storage, and the fabric itself. And you can get to data from any processor in your set of nodes in roughly the same time, within half a percent of the same time as you can the rest of the data. That is game changing. And that allows you to design systems on a much bigger scale than you could ever do before. And design systems to be able to deal with orders of magnitude more data than they ever could before. That's going to take a long time to play out in terms of any fundamental change in the architecture is going to mean a lot of change to the applications, to the stack, everything has to change on the way there. But the endpoint itself is a revolution, in my opinion. Nick, Nick, you helped us start Wikibon. You're back, you just published Storage Key Issues for 2017. And in there, you asked the question, where will your data persist and why? And then you had a statement paraphrasing Gene Amdahl's law, the best I.O. is no I.O. So a lot of what we're talking about today is eliminating I.O. Thoughts on what you heard today? You're really good at squinting through the marketing messages and picking out the weak spots. What do you see as real and what needs work? Well, I'll start off and say bravo to Mike Ron. Somebody had to do it. And they've partnered and done the hard stuff. Now the customers have to beta test. And like you said, they're changing the ecosystem a little bit by moving up the stack. We thought that was inevitable as well. We're not gonna get rid of I.O. I mean, it's hard. So if we could do it all in memory, we would, we can't. This is a pretty darn good alternative. Right, okay. I wanted to ask you guys about some of the other initiatives going on. David, you've had a lot of people debate you about flash, the future of flash, the capacity of the industry to be able to deliver, whether it's memrister, 3D cross-point. What do you see there as far as alternatives to flash? Well, the driving force for flash originated from Steve Jobs taking all the production from Samsung for flash and starting it in the consumer area. And that has been the major driver is the consumer area. So enterprise to a large extent has been on the coattails of the consumer, the business. The challenge of 3D X-Point, which you mentioned, is that there isn't as yet a user uptake of it. And if you don't have those volumes, then bringing the price point down with sufficient volume is gonna, in my view, is gonna be a challenge until there's some sort of capability or some large-scale application which is gonna take use of it. So integrating that into today's architecture, there are some places clearly where it can go. But in terms of volume attempts, other ways of doing it, such as protecting DRAM or DRAM plus flash, are probably going to be used in preference to 3D X-Point. Even as, say, a memory extension or like a persistent expanded storage, even... Well, you'd be able to do that by putting DRAM and then having flash behind it and do a similar sort of thing with the protected DRAM. So you're not optimistic for flash alternatives at this point? At this point. I think they have to come up through the consumer or IoT areas before they get involved. Is this something you've looked at? Do you agree with that? From a pricing standpoint, absolutely. And right now, to really effectively use these coming memories, if you will, you need to put another tier in there and then you need an application that wants that tier and that's pretty problematic. So there's gonna be some bleeding edge users that are gonna adopt this stuff, but without the volume from the consumer, it's just not gonna compete on price. It might compete on performance and might compete in the niches, but not the way flash took the market. You guys heard my comments. I said, I was reading the micron press release and said, centralized pool of storage, improved utilization, improved performance. They didn't mention, I don't think explicitly availability, but they could have. All the things that I said we were supposed to be solving with the storage area networks that really didn't get solved, even though there was a lot of value that was delivered. They solved the problems at that particular time. Yeah, it was so bad, I guess, but it really didn't carry us through certainly the cloud era. And so, and part of the issue was the pace of data growth and maybe it was just architecturally not designed to accommodate all that, but maybe we could start with this whole notion of NVMe and NVMe over fabric to Nick's point, reducing IO with direct access. Maybe you could explain that to the NVMe and then we could talk about the F concept and how likely it is to get adopted and maybe we could project some kind of time frames. So we start with NVMe. NVMe was a direct attach to the PCI bus to a server and that whole protocol is much faster, much less overhead, a retool protocol which replaces the SCSI stack. So that's the first stage is how do I interface at high speed with low latency to the PCI bus in the central processing system. So that's the first systems stage and that allows you what you had back in the pre-sand days which is ultra-fast processors and directly attached to those processors, all the storage directly to that processor. But now you've got all the problems. How do I send a second copy to somewhere else? How do I spread that data with erasure coding? All of those issues are still getting in the way. You've got the same problems that Sam was trying to solve. NVMe over fabric takes the puts in a fabric, a 10 gigabit converged ethernet fabric uses a protocol, rocky protocol, which is RDMA over converged ethernet, uses that protocol to provide very, very low latency, high speed fabric over which you can send that data. And that's the breakthrough here is the overhead of that fabric is down to the five microsecond overhead. So if I'm going locally, they're saying 195, that's gonna come down to maybe 50 microseconds in the next couple of years. The overhead of going outside your particular processor to anywhere else in the network is just five microseconds. So today it's half a percent in the future and that might be five percent or 10 percent. But it's still very, very low compared with the Sam environment and that changes the whole architecture. It enables you to share that data across anywhere and allows any processor to access any data across that whole of that 250. And that's why in that report that you published with the area graph, you see the middle, which is the red, which is the traditional sand just getting squeezed from the hyperscale above and from the enterprise server sand down below. So Nick, in your opinion, how quickly will this be adopted? What are some of the key issues that users need to understand before they can hop onto this stuff? Well, the one thing they didn't talk about today was cost. I don't think we have pricing at all and the Melanox stuff isn't cheap and it's a hundred gig, not 10. Did I say 10? Yeah, well, we all get it all mixed up. Like microseconds, milliseconds, nanoseconds, yeah. Okay, so cost still is huge. The number that stood out to me today was that 10% of the new capacity shipped in 2016 was flash. All right, well, that's gonna be double for 2017, another year or two, it's done. Yeah. Right? We just got replacement capacity and drives are not dying, but the adoption rate, as David says, this is revolutionary. And over this fabric, sands did a lot of what they needed to do. And now this is a high-speed sand with much less latent sand and overhead. And getting that data closer to the processors, you're just gonna process more data more quickly. It's like what they wanna do in the sub-second response time, right? Yeah. So it seems like, well, if you think about it, let's stay on sort of customer implications. I think the start from the top, CIO, CXO, Chief Data Officer as well. I think because it's all about the data, you kind of start with how, data value is really the piece. Yeah. And we heard people talking about data value today, but really understanding how data supports your monetization strategy is kind of to me a starting point for the CIO Chief Data Officer and then really form a partnership with your line of business and then understand what skills you need really to take advantage of this pending change. And that gets into some of the organizational actions. To me, it's all about the developers. We're talking about these new applications that are being developed. I would bring those guys in early to the conversation and then have my architect sit with them and say, okay, how can we better service these individuals? How can we change the workflows? Dramatically improve the productivity of the organization and then figure out all the technology integration activities that I need to do. But I would start with the value of data, seems like far too many organizations, understand that, let alone understand how data supports the monetization strategy. So would you guys add anything to that just in terms of what customers should be doing? So I think the key of this all is that you can do in real time or near real time is of so much higher value than that which you can get to a week later or two weeks later or whatever it is. That's an enormous drop-off in value. So if you can design systems to get any data that's necessary for that system to operate, that is a huge potential increase in value of the data that you have there, dealing with it straight away. So I think that would be one of the principles that I would give to the coders that they or the application designers is go away from your old constraints of I can't get as much data as I want or I can just assume you can get any data that you want and then start with that as your design point. And the second part about it is if you can do your analytics in the same time, I mean obviously there's two levels of analytics, there's the working model and then there's your research model which is improving your working model. So if you can improve that working model as to run in real time with your current systems of record, if you can get those two together, you are going to be able to automate a huge amount of activity which currently goes from the system of record to people to decide things. You can eliminate it and simplify the business processes in a very profound way. And looking for the opportunities that where the most return on that is seems to be obviously the number one thing the CIO should be doing. It was interesting to hear from Micron the value to them was using these analytics to help them with production issues with utilization of capital with productivity, yield from their fabs, et cetera. Those were the issues that are most important to Micron. To other companies it'll be something else. But focusing on the areas of business value and that digital transformation will be, this will allow digital transformation to happen. And I think once you've got that overall concept and you've got people who can think outside the box, obviously that's gonna be one of the biggest difficulties then you can have not straight away but within a few years revolutionary type solutions. Nick, anything you'd add to sort of what customers should be doing? Yeah, they got to get the developers involved early but they also got to get procurement on board earlier because you're gonna accelerate the business which is what you're supposed to do but it's gonna come at a cost, that's right. So get them on board early. Yeah, that's good. Let's see, Linux first, sensible. Block? Well, a lot of this stuff is... Is block, yes. But a lot of this stuff isn't. Well, it isn't, but I mean there is opportunity again and up the stack of combining block and file. So that's value add for Micron's customers, good. For innovators in this space. Oh, you see Pure running hard after file. That's right, that's right. NetApp obviously there. Yeah, yes. So okay, so you don't see any sort of issues there. That's not a big concern. There are a lot of issues which... Webby for Micron. Not for Micron. Well, Micron is going to have to look at these and gonna have to come up with including these architectures for, as you say, block versus file, IoT, edge versus centralized. There's a number of areas there where there's opportunity for innovation. I think the most important thing is to allow other people into this ecosystem so that they can combine and add value together. And that's gonna be one of the challenges of creating that ecosystem. And the power of any architecture, whether it be mainframe or HPE or the, or SAN or whoever it is, producing systems is getting that ecosystem as rich and as open as possible and able to contribute and make money by being in that ecosystem. All right, Nick, we'll give you last word on the day. You've sat through the presentations and what were your big takeaways and any final thoughts? The pace is accelerating. This is, somebody just stepped on the accelerator. Yeah. It's gonna accelerate and accelerate even more and hang on to your hats. Yeah, well, we were early on, I have to say two to our own a bit in the flash and David, you projected sort of the demise of certain segments of spinning disk, but that's sort of less relevant really than the business impact that we've talked about. And of course, by the way, there are people who still continue to challenge those assertions. There's guys like Infinidat that are out there saying, hey, we can smoke all these flash guys so I would challenge Infinidat, okay, we'll prove it. Let's see the benchmarks and let's see the customers and the references. There's a role for this for many, many, many years for magnetic in general, for many years. So old technologies never die, they just fade away. Well, we'll see. Okay, yeah, like Nick said, hang on, stay tuned. All right, that's a wrap everybody. This is theCUBE coverage from Micron Summit. Check out siliconangle.tv, theCUBE.net. Check out cube365.net slash Micron. We've got a site up specific to this event. Obviously, check out wikibon.com and check out siliconangle.com for all the news related to this and other announcements, news of the day. Thanks for watching everybody. This is a wrap, this is theCUBE. We're out from New York City. Thank you. Adhershavak.