 Live from the Oracle Conference Center in Redwood Shores, California. It's theCUBE at the Next Generation Engineered Systems Launch Event. Brought to you by headline sponsor, Oracle. Hey, welcome back everyone. You are here watching Live the Cube in Silicon Valley at Oracle's big launch event. This is theCUBE, our flagship program. When we go out to the events, we extract the civil noise. I'm John Furrier. It might go as Dave Vellante. Our next guest is Juan Luiz, the Senior Vice President of the Systems Group, runs all the software for, I think, one of their important products, database, software. Welcome to theCUBE. Thank you. You're up there giving the keynote today and the messaging is pretty, pretty specific. Highest performance, lowest cost, cheapest gear. Is that kind of an easy window? Lowest administration, easiest to use. So tell us, what's the highlights from now? So give the folks a quick highlights from your perspective you just gave the keynote earlier. Yeah, I talked about our new generation of database machines, particularly Exadata, X5 generation of database machines. And it's one of our biggest releases ever. Includes all new hardware, very fast improvements to the hardware, both on the CPU side, on the networking side. And particularly on the storage side, we have introduced an all flash Exadata that's extraordinarily fast, much faster than anything in the industry. And then I also talked about a lot of new software that we've written. So we have very many specific software features. We've always had a lot of specific software with Exadata. This is our biggest software release ever on Exadata. We've introduced many interesting technologies. We've introduced virtualization in Exadata, columnar flashing in Exadata, fault tolerant in memory processing in Exadata, specialized InfiniBand protocols for OLTP, talking directly to InfiniBand hardware. So there's a lot of different changes. And I also talked about some of the ways that Exadata helps customers save money. So we had John Falleron earlier and Dave made a comment, so we've kicked around before the iPhone for the enterprise. It's been kind of a cliche, but really what you're talking about is integrating software with hardware. John was talking about the integration of the two teams. What is it like now on the advantage side? So performance is significant. You guys are quoting huge numbers in terms of order of magnitude vis-a-vis the competitors. That's going to come from those efficiencies in software. What specifically do you point to in the software? Because the software is where the leverage is. You get a lot of leverage on software. As the power and the scale on the hardware gets tuned in for the software, as an executive in charge of that, what do you hone in on the software strategy? What is the key to success there? What's the secret sauce? You know, the secret sauce, like you mentioned, I mean, we use very advanced hardware and we make it a very powerful scale out architecture, but really a lot of the secret sauce in the software. And why, why do we have this advantage in software? Because in particular for Exadata, we're very focused on database workloads and we control the database software, as well as the OS, as well as the storage, as well as the networking. So we can move functions between these. We can put network protocols directly in our database instead of in the OS. We can put storage protocols directly in our database. We can move database function into storage. So we're able to do many things that you can't do if you just take a database and put it on top of generic hardware, generic network, and generic storage. Because we're able to move functions and we control the entire stack. And that's really the key that makes Exadata different. We also execute very well, but the key is really, we're able to move database functions anywhere we want. We're able to move network functions anywhere we want. We control the whole stack and we're able to integrate it much better than anybody else because we have control of all the software in the entire system. Versus what's been called general purpose, multi-vendor, cobble together the software and then test it yourself, Larry. Allison mentioned that on stage. That's right. That's what you're referring to, right? That's right. The traditional way and the way we've, we still operate and we've operated in the past is, you choose a best-of-breed vendor for database, for OS, for servers, for network, for sand, for storage. And then you kind of put them together and you deal with any issues that come about. This new, the new model, the engineer system model is, we put together the whole system. We optimize it. We heavily optimize it. We heavily test it as a system. And then we go beyond that. We write special software at all the different layers that integrates all of it. And by integrate, I don't just mean kind of make it fit a little better together. It changes. It works differently. So for example, we change the format of data. We don't just move data from disk to flash. As data moves from disk to flash, we reformat the data in a different format. So we're able to do things that you just can't do unless you integrate all the way through all the steps. So Exadata, of course, predated the acquisition of Sun. So you have experience with both models. And you've sort of given us a high-level overview and you've given us some examples. But can we double-click on that a little bit in terms of life before and life after? Because your competitors would say, well, we have a very tight relationship. We work very closely together. We have co-engineering. But you're talking about things that, are you talking about things that aren't possible in that type of relationship? Exactly right. So we do work very closely with many vendors in the industry. Our database runs on practically every hardware platform, storage platform on the market. So we work to optimize to the best of our ability and their ability for their platforms. However, we're constrained in what we can do because it is a platform. It has its own APIs. You can't change the platform. With Exadata, because we owned everything, we were able to basically start from the ground up and say, what do we want? Not what's out there and how do I best put it together. What do we want? And we can design exactly what we want and we can write software exactly for that platform. And we can put functions at any layer that we want. We're also not constrained by standards. So for example, our protocol that the database uses to talk to storage is a custom protocol specifically designed for database. It's not a generic protocol, which are all based on SCSI from 50 years ago. We use very specific, customized, highly optimized protocols. And we're only able to do that because we own all the layers. I want to come back to that. But I have first to have to ask you, Exadata, this is X5, the sixth generation. Did I miss one? Was it, are you not counting the first instantiation? I'm lucky five. That's exactly right. So it's a little bit confusing. It's a minor point, but we started with B1, and then we switched our nomenclature to X, and we switched it to X2. Now really we probably should have said X3, but that wasn't the time. So we have B1, B2, X2, X3, X4, and now X5. So I think it's sixth generation. It's one of those things. It's kind of like lead gear in product releases. Yeah, that's exactly right. It's a little bit like that. We were talking about off camera, about the sort of black area, the white area, and the gray area between hardware and software and engineering. Can you sort of describe those areas and specifically that gray area? John Fowler was on talking about how, sometimes it's difficult to tell who's in which group because of the way you guys work together. Maybe help us understand how you... Yeah, so our software and hardware teams are very tightly integrated. The engineering systems are software led products. So for example, Exadata, Exologic, we are trying to build the best system for the Oracle database, for Oracle Middleware, for Oracle Backup. So it is driven from the software. We have this very powerful software. How do we make the best platform for it? But we work very closely with the hardware team. So when we're planning in Exadata, we have the hardware team and the software team sitting in the same room saying, okay, what do we do? How can we improve the product? We do have different functions. Like for example, there's people that design chips, there's people that design buses, and there's people that implement database software. So we do have specialties. There's not one guy that does the whole thing. But we do work as a team together. We're all in the same room together. We're all designing the next generation together. And you're exploiting things. You mentioned, you talked a lot about Flash today and NVME, and you mentioned SCSI just now. Jung and I often talk about the horrible storage stack. I want to understand exactly what you're doing there and how you're exploiting that. So you're essentially either, are you optimizing the protocol or are you bypassing a protocol and writing directly to an extension of memory? Are you doing atomic writes? Yeah, well what we're doing is taking advantage of Flash the way it's meant to be used. So Flash is a new memory technology. And the way it's initially come into the industry is as a technology that looks like a disk drive. So the easiest way to introduce Flash into the industry was to say, okay, we have this new memory technology. Let's make it look like a disk drive. Because that way I can plug it into an existing machine. Let's go through the horrible storage stack. Right, and that was a good thing because that's how the technology basically got rolling. Adoption-wise. Adoption-wise, that's right. So you took Flash memory, you put it in a box and you put a bunch of protocols in front of it. There's basically SCSI protocols that made it look like a disk drive. And then you could just walk up to a storage array, pop out of disk drive, pop in Flash. Now that's kind of step one of Flash. We've gone beyond step one. We've said, okay, that's fine. That's what we needed to get going. Now, what do we really want Flash to look like? We don't really want Flash to look like a disk drive behind a disk controller because that's gonna limit the performance of Flash. Flash is many, many times faster than disk drives. And disk controllers can't handle that speed. Networks can't handle that speed. So Flash, I talked about how in Exadata we're able to achieve over 260 gigabytes of throughput from Flash. Now that's bytes not bits per second, bytes per second. You can't do that if you try to make Flash look like a disk drive. You can't do that with a conventional SCSI protocols and fiber channel protocols and SAS protocols. They're basically all SCSI protocols. And SCSI is a 50 year old protocol. So we've done two big things with Flash. One is, it's multiple big things with Flash. One is we're using the latest generation of Flash that connects straight into the PCI bus. So it's PCI Flash. That means there's no disk controllers. There's none of that stuff in the middle. So it's the same bus that comes out of the chip, which is a PCI bus. It goes directly to the Flash. There's nothing in the middle. Chip talking directly to Flash. That's one thing. The next thing is the software protocols. So traditionally we've talked SCSI between compute and storage. Well, SCSI's being left behind. There's a new generation of protocols called NVME, non-volatile memory express that says, hey, SCSI was fun. It's been fun for the last 50 years. But guess what? Flash is not SCSI. Flash is not this. It's been a nice ride. It's been a nice ride. It's lasted a long time. It's been great. It's been a nice ride, but it's time to move on. Let's design something specifically for this new generation of storage. And that's what NVME is, non-volatile memory express. You'll start hearing a lot about that. And also the processor advances are right around the corner too. So in memory is only going to get lower latency, right? Right. So you can get lower latency and higher throughput, more IOPS, all of that. And then another part of it is we're using Infiniband instead of FiberChannel, it's a much faster network. Flash is so fast, it's much faster than the network. You have to upgrade your networks. You have to move function into storage in order to take advantage of the full throughput function. It's kind of like 101 here at Silicon Valley. You move fast and then you stop quickly. Yes, exactly. I mean, this bottleneck's going to be somewhere. You move to the next bottleneck. It's great to have the fast car, but if you've got a slow road, it doesn't help you. You've got to make the road fast. So Flash is kind of like a fast car on a slow road. So the bottleneck, you move to the next point of bottleneck, okay? So obviously you're totally, we see that right away. Scuzzy is old, you get a choke point right there. What's the strategy and what's the debates internally? I mean, I'm sure up and down, because you're building a platform. You're not just tackling storage in a silo. You're not in a vacuum on that. You're looking at a holistic platform. So what is the next bottleneck and what are you guys engineering that around and how does that road map look? What can you share, some insight into that direction? In memory, database? Yeah, so the other big thing that we're working on is in memory processing. And there's a lot of work. We just released our in-memory database last July. We've done a huge amount of work on not just moving data into memory, but optimizing the database and the formats for in-memory processing. So we have special columnar formats. We have special vector processing. So we actually use the vector processing on the CPU chips. Those are instructions that were put there for graphics operations and for scientific computation. We're actually using those in our in-memory database to accelerate data processing. So we've done a lot of integration with special chip functionality and you're going to see more of that from us. So as time goes on, we're specializing our hardware, also our chips, to accelerate in-memory processing. So you'll be hearing a lot more about that in the near future. So software-led infrastructure, software-led design, Wikibon really put out some of the first, actually the first research on software-led infrastructure, but it was IO-centric, was the kind of the first initial word we talked about. So that is a network issue. So where's the network fit in? Because you have mentions of Cisco, you mentioned Cisco on stage, Larry did. So they're a network provider. So when you do fully engineered system, you got to address the network issue. So up and down the stack, where is that software improvement that you guys see with this announcement that's most compelling? Storage, network quality above, what's the key? We've improved everything. We really improved everything. In the case of networking, what we're using is the latest, most modern networking technology, which is InfiniBand. Again, Ethernet's been around for a long time. It's a very good technology. It's actually been enhanced quite a lot over the years, but InfiniBand is a newer, more modern technology. It's faster, lower latency, higher throughput. That's what we use. And on top of that hardware, we've built special protocols. So we don't just use the hardware, we write special protocols on it that are specialized for our database functions that make it run even faster. So for example, we can now talk directly from our database to the InfiniBand hardware without going through the OS at all. And what's on the other side of InfiniBand might be whatever, right? You can still have the Exadata in here. That's right. Pointing big fat pipe to something else. Yes. I wrote a piece four years ago, I think it was entitled Larry Loves InfiniBand. You guys love InfiniBand and we're now starting to see the fruits of that. Well, you know, it's not so much that we love InfiniBand, it's that InfiniBand works so well. You love it for that. And the other part of it is extremely reliable. We have, you know, it's very impressive to call it because despite being so fast, sometimes you think it's like this super fast car, it's going to be kind of flaky. It's rock solid. It's actually more reliable than Ethernet. It has more, it has guaranteed packet delivery. It has better latencies, guaranteed latencies. It's a more reliable protocol also. You were listing off a number of innovations. I couldn't keep up. InfiniBand, virtualized InfiniBand, JSON data types, JSON analytics, offload, some high availability stuff, fast IO failover. That's all software related, right? So that's in your sort of direct purview. Is that correct? And there was a long list of other stuff, direct wire to InfiniBand. We had a customer last week, Big Bank saying that they were going to take their mainframe connected through InfiniBand to Exadata to bring their analytics and their transaction systems together. Are you seeing sort of more of that type of trend? Yeah, so InfiniBand, we were the first major enterprise vendor to adopt InfiniBand. And now you're seeing a lot more of that. So Teradata has now adopted it. IBM has adopted it. So a number of vendors are starting. EMC is now in their latest flash stores. They use kind of a combination of fiber channel and InfiniBand. You're seeing this in industry. We were probably, we did it seven years ago, five years later, a lot of other guys are saying, well, they were kind of right. We'd better do this also or else we're going to be left behind. You've also bundled in virtualization to some of your solutions. Can you talk about that? Why now? Why not offer a choice? Everybody talks about choice. Why your selection to bundle it in? Yeah, so we've introduced virtualization into Exadata. It's based on our OVM Oracle Virtual Machine, which is a Zen-based virtualization. This took us a long time. It was a big engineering effort. Virtualization has existed for a long time in the ethernet world. So a lot, a lot of engineering has gone into virtualizing ethernet networks. And that's been done by thousands of people throughout the industry. On the InfiniBand side, we are the very first vendor to introduce enterprise class platforms that are virtualized using InfiniBand. And we had to do the heavy lifting ourselves. So we had to write a lot of software. We did a lot of work so that we can combine the two things, we can combine the performance and reliability of InfiniBand with the flexibility of virtualization. So that's the big effort that we've done. And we've married the two and we've done, we've written all the software all through all the components with our partners also that made that possible. I was going to ask you, so what happens to that? I mean, is it, do you keep it inside Oracle? Do you share it with partners? Do you share it with the industry? Do you open source that? I mean, what's on that spectrum of how you share that? What do you share? Yeah, so in terms of our OS and virtualization, we use Linux and Zen and we share everything we do on Linux and Zen. So at that level, everything is shared. In the database level, we have a lot of special protocols. Those are specific to our database products and they're built into our database and those are essentially proprietary to our database. So it's kind of a mix. The platform level, everything's shared. Everything's open source. At the database level, everything's customized. Can you explain the new elastic configurations and kind of what that means? Yeah, it's pretty straightforward. So exadata in the past, we've had fixed sizes. So we had something like called a quarter rack, a half rack and a full rack and you move between these sizes. And the nice thing about that was we were able to carefully balance compute and storage. With our new generation of machines, we have something called elastic configurations which is simple, which is you start with a small exadata and then you're able to add computer storage incrementally one server at a time as you need it. So it's much more flexible. It's a buy as you grow kind of thing with the data. Exactly, buy as you grow, smaller units of expansion, expand just the storage or just the compute or add just flash or just high capacity storage. We just made it much more customizable, much more flexible and because of that, also much lower costs. So the thing about what it was different before was it was prefixed, prefabricated units which was optimized obviously. So the only thing that does differently allows customers to get into the product in different use cases. That's right, you can grow in smaller increments. Also you can customize an exadata. For example, if you're running an in memory workload, you don't need that much storage. What you want is a lot more memory. Got it, okay. So you can grow just that part of it and shrink the other part to basically get the right balance of compute memory and storage. So you're going to see more use cases then expand out for that. Yes, in fact, we did that for exactly. The driving use case for this, it was really two driving use cases. One was in memory database. So this is something that's starting to take off now. We just introduced our product in July of last year. And the other use case is all flash storage. So that's another important trend in the industry with all flash storage. You know, our all flash storage is different from other people's because I mentioned we do a lot of special stuff with our flash storage. It makes it much, much faster and more effective. There's been some comments coming up on the crowd chat and the Twitter sphere around ecosystem, you know, and that Oracle, you become like Intel. You're going to have all the chips on your table, which is normal land grab. We hear that comment, oh, the land grab. You guys have a partner network. You have an ecosystem. How does the ecosystem change in this new world? Because, I mean, I think no one's denying integrated stacks and engineered systems as a way to go. It's clearly, it's got the vote of confidence from the industry. You see a lot of the high performance stuff going that way for workloads and whatnot. But how does the ecosystem balance of power change? Because you still need partners. Now the OS might be a bigger part of it. You're seeing operating system kind of configurations come in. How does that change the ecosystem, if any? Or does it? Well, you know, there's just kind of two parts to that. There's the ecosystem that uses the Oracle database. So we have a ton of ISVs that use our product, that build products on top of ours, that build tools on top of our product. All that's completely unchanged. So all those partners we still have. And in fact, it's much easier for them now because they don't have to worry about low level issues that are messing up their product. So if you're building, for example, a telecom application on top of Oracle, before you had to worry about, well, is the customer going to deploy it on the right hardware? Is it going to tune it right? Is it going to mess it up? Because they're going to be blamed for it as the guy on top. You do the integration work for them, basically. So we've done all that integration for them. So you basically can't mess it up. And not only can you not mess it up, you get a really best-in-class integration. Sounds like the perfect IT solution. You can't mess it up. But so they add services on top of this. With them, it's all gross profit. They don't have the cost-carrying on the integration. Yeah, it makes it simpler, faster, better, more reliable. So ISVs, what about other partners? In terms of other partners at the hardware level, we still have all that. So Oracle continues to support any hardware platform you want to run our software on. So we still run on every major server vendor, storage vendor, all that still goes on. We still optimize on those systems. We also have in parallel our engineered system. So it's an alternative to, it's basically another way of deploying Oracle. We're here one of the ways. The Senior Vice President of the Systems Group, really the heart and soul of the kernel of the crown jewel exadata, which is driving the software-led innovations across the company. Thanks for coming on theCUBE. I'll let you have the final words. Share with the folks out there in your own words. This announcement, what's this all about today? This whole big deal. Yeah, the big deal is our new exadata platform. It really is our biggest release in many, many years, both on the hardware and the software side. We've introduced a lot of new technologies and new flexibilities, both on the hardware and the software side. We've introduced all flash storage in memory configurations for exadata, elastic configurations, virtualization, snapshots, specialized protocols, specialized caching for virtualization, for analytics. There's a whole slew of new technologies. It's hard to sum it up because there's just so much. And that's actually one of my challenges is there's so much that's out there that we've introduced that it's hard to actually summarize it in a few words. Let me guess you're not done. There's a lot more coming. And so, yeah, so customers of exadata are going to be very pleased with this. They're going to see huge improvements. And it's only adding to the differentiation. So as time goes on, we keep adding more and more software functionality. It further differentiates the product. It makes it better for us. Well, there's certainly interest. We saw a huge live feed commentary of people watching the feed after they're interested in your moves. It affects their business and certainly any innovation that creates more and more opportunities and lower costs, higher performance. I mean, that's a tried and true formula for success. Higher performance to the lowest cost possible. Congratulations. We are here live. We'll be right back. This is theCUBE. We're here with Dave Vellante, live in Silicon Valley at Oracle's headquarters. We'll be right back after this short break.