 Okay, we're back. This is Dave Vellante. We're here at 590 Madison Avenue in New York City at the big IBM announcement today. Flash ahead. I'm here with Vincent Sue, who's an IBM fellow and CTO of the storage business. Vincent, welcome back to theCUBE. Thank you. Good to see you again. So, big announcement today. You guys are all in on Flash. Talk about Flash as a game changer. Why is it such a game changer from a technical perspective? You know, over the last, you know, 50 years, the whole system design and the software applications are written pretty much to start to accommodate the IO problem. As you can see that for the last, you know, 10 years, the CPU has improved 10 times, they were improved 100 times, but storage three performance wise only improved 1.5, 1.2 times. A lot of the system architectures and software are really trying to work around their problems. Now, when the IO bottleneck is removed, okay? You, all of a sudden, a lot of things become possible. Okay? Not just the traditional application. Now we can do more, because we can do more with a much shorter time. Yeah, so, on the one hand, we heard from customers today that they're basically dropping in an all flash array. It's a block-based device. Looks like any block-based device and it immediately accelerates performance. On the other hand, over time, there's great potential to sort of re-architect infrastructure and applications. What do you see as that potential and how will that change infrastructure, applications, and ultimately business? Yeah, so for example, today, let me just give you one of the most obvious examples. Today, in the database design, we, everybody do database reorganization, database re-or, right? You have to re-or database-wide because you want to get the best performance so you need the data sort of adjacent to each other on the plan, okay? We need to question, those business probably doesn't need to do it anymore or doesn't need to do as often anymore. So all those parameters that we do to tune our system, those had to be re-think, even in our IO path, okay? Today, we try to buffer up as much as possible because go to states is very slow, right? How much metadata you can put in the system to make the sense, so allow you to make sense from the data. Those things in the past is very hard to do because every little IO goal to this is very slow. But now, because the Flash-enabled system allow you to put a much better metadata-rich systems available so you can do a lot more with what you have. So let's talk about metadata a little bit. So metadata today is sort of locked in the device, whether it's a network device, a server, an array, and it's sort of controlled by that device. Do you see that changing? Will this whole notion of software-led or software-defined infrastructure allow metadata to actually be a shared resource? First of all, first question, and second question is, where is that going to get managed? Okay, so you're absolutely correct that the metadata is where, first of all, I mean, metadata will allow you to make sense from the raw data you have. So I think that we are getting to the point that this world is going to have a very, very metadata-rich information, okay? But in the past, because the metadata is, you know, compared to the raw data, metadata is much smaller. You know, you go to those smaller IO, it costs you a lot. So a lot of people that buffer up the metadata in the DRAM memory. So it's sort of limited amount of data metadata you can have. But these days, now with the Flash systems available, you're able to do a much better-rich metadata and be able to share it between the different applications. The thing is that these days, the amount of time that we see over and over again, different applications have the same content, but they cannot really share the data without reshaping their metadata around the raw data. So now with this kind of technology, we enable you to be able to share those content much easier once you can transform those metadata quicker. Yes, so you heard in my panel that I was hosting, I was doing some back-to-the-napkin calculations last night and I had, you know, CPU speeds at nanoseconds, 10 to the minus nine, and disk speeds, 10 to the minus three, six orders of magnitude, Delta. I'm not sure if that's exactly right, but it's big, right, close enough, right? So my question then is, who's going to control that metadata? You can't really control that metadata from slow storage. Doesn't it have to be controlled from fast servers? In the past, people tried to put automated in the DRAMs because if you control a metadata in a slow storage, then it's just too slow, right? It's just, especially the metadata is a very small IO. Small IO is extremely expensive to the hard disk, but the problem is if you want to have a large amount of metadata to make your data, to be able to make sense from your data, that you cannot put everything in DRAM, and that's where the flash or flash system come into play here. Yeah, excellent. So how do you see this thing progressing? You guys are putting a billion dollars in. Obviously, you've got investments organic and potentially inorganic. We're going to ask Steve Mills about that, but where do you see this whole flash thing playing out over time? Okay, so the flash optimization is across all the layers from the fundamental core technology. That's what the Texas Memory System has done a superb job to make the flash more durable and high performance. Then we have an overall system design, the power system, Z system, the storage systems. How do I design the system? So, the rest of the system is not going to be the bottleneck. Then we have a middleware design to take advantage of that. Remember I talked about the database rewars and there are a lot of, the software defines storage interface to allow the middleware's application to be able to truly take advantage of those technology because at the end of the day, people don't want to spend their time defining where the data placement is going to be. They want to be able to talk to these devices. At the end of the day, what IBM vision is going to be, take this storage problem from the application. All you see is an infinite flat space. You can just rewrite to it and everything will be persistent. Back to the days of single level store and it's super fast. Vincent Sue, thanks very much for coming back on theCUBE and great to hear your perspective. So keep it right there, everybody. We'll be right back with our next guest.