 Okay, we're back. This is Dave Vellante. This is the special FlashCube, siliconangle.tv's continuous coverage of the biggest market trends we can find. Flash has taken the world by storm, but it's not just about the hardware. It's really about how database design is changing and how that's affecting business performance. And we're here with Brian Bukowski, who is the CEO of a company called Aerospike. I first met Aerospike through John Furrier at Oracle OpenWorld. It was astounded by some of the very interesting things they've been doing. We've talked to a number of their customers. I'm here with my co-host for the moment is David Floyer and Brian Bukowski. Great to talk to you again. Hi, David. So Brian, we've had you on theCUBE a couple of times now and really appreciate you coming to the studio at Siliconangle. And I wanted to start by just giving us a quick update for those who don't know Aerospike, no SQL database, you're all about like super mega fast performance. You're leveraging Flash in really innovative ways, but give us an overview of the company and tell us what's new. Sure, so Aerospike was funded four years ago and venture backed company here in sunny Mountain View, California. One of our goals is to take some of the lessons of direct attached storage used by some of the folks working at the highest levels of scale. So we know that Facebook and Apple combined to buy $200 million worth of Fusion IO hardware last year, that's in their public reports. They must be doing something with it. They're using direct attached methodologies. They're changing the kind of databases they use. At Aerospike, our goal is to take that kind of technology in that kind of scale and apply it to a broader market, allow anyone who doesn't have the ability to build a hundred person engineering Google style staff to apply these kinds of solutions to their businesses. It means being able to quickly put together any clusters of between one terabyte and a hundred terabytes with a screaming fast flash underneath it and not just get the two X to four X you'd get by putting a relational system on top. With at Aerospike, we've focused on flash and flash and DRAM is a hybrid technology for real time and stream oriented data. So Brian, talk about this big theme around in memory databases today. Intuitively, memories faster than flash is faster than spinning disk. Talk about why flash as a memory extension as opposed to an in memory database. Yeah, interesting thing. I was talking to one of the key database analysts at Gartner just last week and he said flash is storage, flash is not memory. In memory computing is DRAM, it's not flash. And I had a chance to evangelize upon him what we think at Aerospike and really changed his mind on the topic. And here's the difference. If you use the standard software techniques of database indexes and really just think of flash as storage, yeah, it does only go two times faster. But there's a different way to think about flash and a different way to access flash. At Aerospike, what we did was we said it's the index updates that are hard. That's where you have to continually write and you have to have things perfectly in sync with all the different parts of storage. That's what makes index updates and writes and reads difficult for classic databases when thinking about it as storage. Memory is all about random access. Well, there are techniques, there are indexes, there are ways of writing that software that allows you to really unlock the true speed of flash. What we've been seeing in the marketplace over the last just six months has been pretty incredible in that there are devices that have gone from 50,000 IOPS up through 500,000 IOPS on individual PCIe cards. We're seeing SATA being left in the dust by some of the new PCIe cards and the kind of revolution we were expecting where flash performance gets a lot closer to DRAM and really the two of them are a pair in terms of what the capabilities are. All of that's happening right now. So, David Floyer, we were talking at the top of the hour about database design and you talked a little bit about how databases are not really that flat today. Can you talk, David Floyer, specifically about the flattening of the database infrastructure, bringing together transactional and analytic systems and then tie that in with some comments and questions to Brian on what that means. Sure, and I'm interested in some of the examples of your customers, what they're doing in this way. Does that model that I was describing, is that the way they're thinking about things, trying to get things done in very short periods of time, flattening the database design, being able to adapt to a new environment very quickly. Can you talk about that? Sure, sure. I think that's a coming trend. I'm not sure it's quite here yet. The world of having a front edge database, a real time system, as well as having a true analytic system, we're seeing that in our customer base as still getting closer, but it's still apart. One of our customers is a company called App Nexus. They're one of the largest advertising platform companies in the world. They're second to Google and probably that's about it. And the CTO has a great talk at Surge last year where he said, you know what, we've got all these different analytics engines because we have all these different queries. Every single one of them is optimized for a different thing. When we've got six petabytes, and I can't remember how many petabytes they're accumulating a day, and we want to do big table scans, of course we use Hadoop, that's the tool for the job. They have a Vertica cluster. They use that on the faster, tighter sets of data. And then he said, okay, great, you've got all those analytics pieces because all the special purpose forms of analysis, what do you do on the front side of that? You really only have one database that is interactive that's on the front side of your service when you're building a service architecture. For that, we use Aerospike. So we're still seeing a lot of separation. I get a lot of resonance when I tell people, Aerospike customers about how they can keep their front edge database and then have a real time connector over to an Aerospike analytics system and have them separate. And they feel comfortable with that still, where there's this move between the two systems. Their real time system is staying real time. So I think that's gonna be with us for a little while. Now as the systems get faster and faster, there will be this natural collapsing of the infrastructure. On the other hand, you're doing an analytics job, it's hard to do a time limited analytics job. How many rows are you gonna touch? So some of that technology will be coming with us, but right now a few people still feeling comfortable with separation. Okay, so but do you see long term, you said the reduction in time from the technology there. Are you seeing that continue down that line and can it improve? Absolutely, and here's why. I was at a conference called the Merchant Risk Council talking to fraud and risk analysis companies. And the first guy I talked to, he was at a particular company that does a lot of transactions. And the first thing he said was, how do I ETL into your database? He didn't even care what it did. As long as he knew he could get his data into it or couldn't get a data into it, that was gonna be the make or break. Then we talk about what power your database had over other databases. So ETL and the problems of ETL and moving the data back and forth foremost in his mind and many people's minds. So if we can get more powerful databases and more individual systems and start overcoming some of this special purpose-ness, the special purpose-ness has been great because you find a business problem, you find a set of analysis, you can get that database that does that 10 times faster. Now you've got this data motion problem you're all trying to link. So the pendulum is swinging back and forth still. I think it's gonna come back in the sense of having a fewer platforms and less ETL because everyone's worried about ETL now. Okay, so one of the areas you might tackle at is by that use of metadata of one source or another know where that data is and be able to extract it and bring it back in faster. How are you designing that into your products or your future products in order to tackle that whole problem? Sure, so at Aerospike we specialize in some sense in smaller chunks of data and making those chunks less localized. So there's a one way to do it is to try to say tackle metadata is put your metadata with your data and then you've already got the data right there if you wanted to look at it. We actually look at the problem a little differently because you almost certainly want that data to be in another store, right? So if you've got a very fast metadata store that really is distributed, then you can put your metadata somewhere else and just say it tells someone, okay, go get it over here. So by focusing on the problem of small fast data you can build out a data architecture. Maybe it looks like the true world of MDM from the financial services industry 20 years ago and the kind of vision that they had of a metadata store really pointing at. That's really a great use of Aerospike as well. So I wonder if I could ask you, Brian, so what kind of data are we talking about here? You know, you think of databases, you think of structured data, you think of flash, you think fast. Everybody's sort of focused on that database problem. Are we talking about structured data, unstructured data, semi-structured data, both, can you talk about that a little bit? Sure, so I think the term structured data and unstructured data is almost gonna vanish in the next five years, that's a prediction. Because what we really have these days is structured data was required because of tabular formats for rotational disk. That's why you had to have a strict number of columns and that restriction is vanishing rapidly day by day. So what we still have is you need structure in that you need to know what data is in a given row, what's in a given table. You need enough metadata to know the meaning of what's in there. So schemas and a hint of your data, the ability to determine what's there, that will stay with us. And in some sense, that's part of structure. That says this data has these fields and they mean this. That'll be with us. This idea that structured data means needing to add columns and having particular performance problems associated with us, that's with the type of no SQL stores like Aerospike already passed because we're not dealing with rotational disk optimizations anymore. So we'll have structure, you need structure, you need to know what your data means. On the other hand, adding columns is also a thing in the past. Excellent, Brian. Well listen, I wanna thank you for coming in on such short notice. You're always a great guest and love to have you on. Great innovator, Brian Bukowski, CTO founder of Aerospike. Really appreciate you coming on theCUBE. Keep it right there, buddy. We'll be back with our next segment. We're gonna bring back John Furrier. We're gonna riff on what we just heard. Set up the next segment. This is Dave Vellante. This is theCUBE. This is the FlashCube. FlashAhead, we'll be back right after this commercial. We looked at all the programs out there and identified a gap in tech news coverage. There are plenty of tech shows that provide new gadgets and talk about the latest in gaming. But those shows are just the tip of the iceberg and we're here for the deep dive. There's a difference between technology consumers and those who live the business day today. And our viewers recognize that. The market begged for a program to fill that void. We're not just touting off headlines. Our goal is to provide you with a story, but we also wanna analyze the big picture and ask the questions that no one else is asking. Our guests aren't just here to provide commentary. We work with analysts who know the industry from the inside out. The tech business isn't new, but many networks treat it as if it is and really barely scratch the surface on technology coverage. We follow the expansion of the cloud and the evolution of big data. We're covering new enterprise from startup to IPO and every move in between. So what do you think was the source of this misinformation? And so you mentioned briefly, there are several other. If that's the case to players. I like to think of us as a companion to theCUBE. We're here every morning trying to.