 Okay, here we're back live at Velocity Conference. This is Wikibon and SiliconANGLE. Go to siliconangle.com or wikibon.org because you're going to find the blog post from this conference at siliconangle.com and Wikibon has all the summary reports of all the great content we're producing from Velocity, about 17 interviews a day, a lot of editorial content, a lot of great entrepreneurs, amazing content. This is theCUBE, our flagship program. We go out to the events. You can try to see them from the noise. I'm John Furrier, the founder of SiliconANGLE. I'm joined by co-host. I'm Kay Vellante at Wikibon.org. Srinivasan this year. He's the co-founder and VP of engineering and operations at Aerospike. Aerospike's a company that theCUBE first introduced you to last year at Oracle Open World. We had Brian Bacowsky on. We talked about what Aerospike is doing. Subsequently, in the Wikibon community, we've had some Aerospike customers that we've brought on to peer insights, namely TAPAD is a company that's a really interesting case study. Srinivasan, welcome to theCUBE. Glad to have you here. Thank you, Dave. It's a pleasure to be here. Yeah, well, I appreciate you making the time here. This is a conference about web performance. You guys are all about performance, aren't you? Yes, we are definitely the fastest database out there in the NoSQL space. And are able to generate, you know, hundreds of thousands of transactions per second at sub millisecond response times. And we have basically been working on speeding up our applications, you know? Yeah, so there's a lot of talk here about, you know, a lot of question marks. Is the web getting faster? You know, are networks getting faster? Are page loads getting faster? Is this, you know, server performance improving? Does cloud help or does cloud add complexities? But, you know, what's your take on just web performance in general, and specifically, what's Aerospike doing to speed up the web? Yeah, my take is based on my experience earlier from a large internet company, you know, Yahoo, in fact, where we ran very large, you know, web-scale services. The biggest user demand was for immediate, you know, vitality on your applications. So that is actually what is driving, in my opinion, the web performance improvement, is people really want to see the latest and greatest information, and they're not prepared to wait. And this is especially true in the area of real-time advertising, where Aerospike kind of cut its teeth, if you will, where we needed to deliver day-in and day-out performance for real-time bidding. The entire process was within 50 milliseconds, and in that time, people had to do, you know, compare multiple display advertising pieces, and what to show to a user. And if you can't actually show all of that within a short amount of time, the user is gone, so you don't actually make a sale. So you have to run all the time really fast if you don't run and complete these requests within very short, you know, five milliseconds or two millisecond response times in some cases, you don't actually have any business, and you have to run all the time. Yeah, so that's why speed is so important, right? You're talking about business value, right? You lose customers. Exactly, right. Exactly, and in fact, the interesting thing is if you look at the internet and the way customers go away, you know, downtime is not an option at all, right? I mean, that's kind of true. What is kind of, people are kind of fighting over is how many people will stay for one second, versus two seconds, versus three seconds? I think there are some case studies, I think the large internet companies have done, where if they lose, even if they slow down by 20%, they lose a lot of users. Yeah, so you talk a lot about many of the cube events that we go to, there's not much of a discussion here, although potentially there should be about flash and how that changes application development. You guys are in the heart of that, using flashes and extension of memory to a much more dramatically speed up performance. Talk about that a little bit in terms of specifically what you're just saying, you know. Is, you know, three seconds versus two seconds versus one second, does it really make a difference? The difference it makes is not in an absolute sense, but in a relative sense. If you had two services on which one of them could actually return to you consistently in one second with a better user experience, because they are able to crunch all this data in a very short amount of time and give something relevant to the user, I can tell you that the majority of the users will actually end up on that one millisecond site. There's actually big precedence on this, okay? There was once, I don't remember the exact, I think it was Jim Barksdale, he actually laid cable from Chicago to New York and then because of the curvature of the earth, they could actually shave off a few milliseconds out of the trading. So very soon all of the commodity trading in Chicago had to actually shift completely en masse to the new cable system that Jim had actually put out. And that's usually how it works. The competition works that way. If you're better, like you said in your case, the one second one is going to win 100% of the business and not, you know. I didn't know that story, Barksdale was always the leading edge here. So okay, so let's talk a little bit more about sort of three things, speed, scale, and reliability. You architected this NoSQL database really for those three attributes. Talk about that a little bit and what kind of proof points can you share with us? Yeah, the interesting thing about speed is I think a lot of people do not worry about vertical scale. Now there is Moore's law on your side. If you want to basically exploit the multi-core technology that's been invented and pretty much deployed everywhere and also the network cards, everything is improving in terms of hardware. You want to actually squeeze the best performance out of every hardware. And this is the way we look at software. We write our software in C. We stay as close as possible to the hardware and make sure that we get out of the way of the hardware so you can get all the performance that the new hardware brings into your application. That is vertical scale. So we're very, very, you know, kind of, that is the speed part, okay. And then what happens is there is a scale part which means put together multiple such high performance nodes, if you will, together and form a cluster. What that allows you to do is enable people to add capacity on the fly to be able to make sure that your service keeps running when you have to do these kinds of things. You know, when a node goes away and so on, you get the level of resiliency. And then it never fails is even more important because what you do is the self-management of the software keeps the service running all the time with an SLA. And why does the speed help? The speed helps because you have so much extra capacity in the system and this extra capacity essentially enables you to handle unforeseen operations that come into play. When you have a lot of data in the system and you're doing reads and writes across, you know, what your applications have to do which is basically behind the applications of the users who are the people who react to the one second versus three second differences, right. That's something which is going on all the time. Sometime a bad thing happens like a node dies, you know. At the point when a node dies, you still want your users not to get a three second response time. So this extra capacity which is there in your system enables you to tide over by, for example, in the background you have enough capacity to rebalance the cluster's data while these high performance transactions are going in and out. And you really don't have, you don't skip a beat, so to speak. And that's what all these things help you to do is to run a service 24 by seven at high throughput and low latency. Okay, but the complexity at that scale starts to get greater and greater and greater. How do you manage all that complexity? That's actually a great question Dave because one of the fundamental technological decisions that we made as we founded our company, you know, as Brian and I started discussing this, was to keep our architecture extremely simple. Every one of our nodes is self managing but it's also the same as every other node. There are no classes of things which people have to manage. And the hard part, no hierarchy, the hard part is actually in the software where each node had to deal with all these hard issues because it had to do not just simple transactions, it had to deal with SSDs, it has to deal with the data rebalancing, it has to deal with defragmenting the disk. All of that is done in every node. So the code is hard but from an operational point of view it makes it very easy to understand the system and manage it. And that's important because we have large clusters which are running 24 by seven with virtually unattended, 90% of the time and 99% of the time because only when you want to upgrade something you go do some kind of manual intervention. And even there it's all automated to the level possible. Trini, I want to ask you about the company AeroSpike, I see Brian's good friend of ours, been on theCUBE before. And we knew you guys when you were Citrus Leaf before. You guys are a bunch of alpha geeks, we know that over there. So we know you guys are doing some good engineering. But give us an update on the company, where are you guys at right now? You guys are a lot of momentum. The market really spun in your direction, certainly the in-memory database, you guys were way ahead of the marketplace on that. What's going on with the company, the engineering team and the product? I mean, the short answer is we're growing in all aspects. Our customer base is growing. We are starting to now move more into different kinds of verticals. The interesting thing about speed and scale is while there was a lot of kind of uptake in the real-time advertising market, we're now moving into other verticals like retail, financial services, gaming and so on. So we're kind of expanding our footprint. We're also expanding our product. In terms of engineering, we have teams in Silicon Valley and also in Bangalore and we run a 24 by 7 shop. We are building the next version of the product. Our basic strategy is to build a database company pretty much the way the early database companies have been built. Our advisors are all stalwarts from the old database world. There's Don Hatterley from DB2, the father of DB2, that is Roger Cipo, founder of Informix. There's Marianne Nimet, who basically ran TimeStand. So we are basically building the database company. They could be holding you back. They're old school. Is it old school? Is it new school? Well, the important thing is we believe in, Brian and I are definitely new school. I'm only kidding, by the way, those guys are legends. If you've ever met Don Hatterley, you would know that he is a world class in not just smart, but he's actually forward thinking to a level I think all of us can learn from. I was kind of joking there, but it's good to have those kind of advisors because the database, we just talked about the young startup here that was out early that's coming out from the East Coast. It's a couple of young guys and it's cool to be database guy again. I mean. Yes, and that's the fun part for me. So I've always, after my PhD, I've always wanted to build a database from scratch and this is my chance. And the timing is perfect with big data and everything going on, it's great. Well, hey, thanks for coming on theCUBE. We're a big fan. But I got one question. We're going to ask you, because John's right, you guys are alpha geeks. So one of the use cases for Aerospike is AdServing. And my question is how fast can it actually get? We saw this technology coming out, pre-persisting data. It's like taking streams and trying to make decisions. I wonder if you could give us opinion on that, have you seen that or how you can take advantage of that type of technology? Well, one of the things about streaming is you can obviously process data as it comes in really fast. But what if you wanted to actually correlate it with information or knowledge that you have gotten maybe five minutes ago or one minute ago. Or a year ago. Or a year ago or even longer. That's where something like Aerospike comes in. So we would work in conjunction with something like streaming. So we can actually keep up with it. You have a database in Aerospike which can keep up with extremely fast data rates. So you can merge the streaming kind of processing and compare it with historical data in real time. So you know exactly at the same point that you're observing something. What exactly it is? You can compare it to what was observed earlier. So do you see that as sort of a next wave? That is actually a huge thing we are actually working on. For example, there are things like Storm and so on which I think are a great match for the database of Aerospike. Great, I just wanted to get your opinion on that. All right, Serenie, thanks very much. Appreciate you coming by theCUBE. All right guys, thanks for watching. We'll be right back with our next guest. We're live at the Velocity Conference. This is theCUBE, our flagship program. We go out to the events instructor to see them from the noise. All the videos will be on youtube.com slash siliconangle and go to siliconangle.com. We have blog post covering Velocity and for other tech, go to wikibond.org. You have all the great research there and all the summary pages on the wiki from Velocity Conference. Again, this is where the action is. This conference really is bringing together infrastructure, cloud ops, dev ops with development on the front end. That's the convergence on top of the stack. That's mobile, that's web, that's performance and underneath it's big data and memory databases and the stuff we love. So this is theCUBE. We'll be right back with our next guest after this short break.