 Live from the Sands Convention Center in Las Vegas, Nevada, it's theCUBE at AWS ReInvent 2014. Brought to you by headline sponsors Amazon and Trend Micro. Okay, welcome back everyone. We are here live at Amazon Webster's ReInvent 2014. This is theCUBE, our flagship program, where we go out to the events and extract the synth and the noise. I'm John Furrier, the founder of Silicon Am. I'm John Michael, Stu Miniman from Wikibon.org. Our next guest is James Hamilton, who's Vice President and Distinguished Engineer at Amazon Web Services. Back again, second year in a row. He's a celebrity, everyone wants his autographs, felpies. I just tweeted a picture of Stu. Welcome back. Thank you very much. I can't believe this is a technology conference. He was falling over himself right now because he's so happy you're here. And we are too, because we really appreciate you taking the time to come on. I know you're super busy, you've got sessions, but I always could do a CUBE session on, kind of what you're working on. Certainly, amazing progress. We were really impressed with what you guys have done over the years last year too, but this year the house was packed. Your talk was very well received. Cool. Every VC that I know in enterprise is here. And they're not telling everyone. There's a lot of stuff going on. The competitors are here. And you're up there in a holding court. Talking about the future. So, quickly summarize what you talked about in your session on the first day. What was the premise? What was the talk's objective? And what was some of the key content? Gotcha, gotcha. My big objective was, the cloud really is fundamentally different. This is not another little bit of nomenclature. This is something that's fundamentally different. It's going to change the way our industry operates. And what I wanted to do is to step through a bunch of examples of innovations and show how this really is different from how IT has been done for years and gone by. So the data center, obviously, we're getting quotes after quotes. Obviously, we're here at the Amazon show. So the quotes tend to be skewed towards this statement. But I'm not in the data center business. Seems to be the theme. People generally aren't in the data center business. They're doing a lot of other things. They need data centers to run their business. With that in mind, what are the new innovations that you see coming off that you're working on, that you have in place, that are going to be the enabler for this new data center in the cloud. So that customers can say, hey, I just want to get all this baggage off my back. I just want to run my business agile and effectively. Is it the equipment? Is it the softwares or the chips? And what are you doing there from an innovation standpoint? Yeah, what I focused on this year, and I think it's a couple important areas are networking because there's big cost problems in networking. And we've done a lot of work in that area that we think is going to help customers a lot. The second one is database because databases, they're complicated. They're the core of all applications. When applications run into trouble, typically it's the database at the core of it. So those are the two areas I covered and I think that's two of the most important areas we're working right now. Yeah, so James, we've looked back into people that have tried to do this services angle before. Networking has been one of the bottlenecks. I think one of the reasons why the XSBs fell to the 90s, it was networking and security, grid computing. Even to today, so what is Amazon fundamentally doing different today? Why now is it acceptable that you can deliver services around the world from your environment? What's different about networking today? It's a good question. I think it's a combination of private links between all of the regions. Every major region is privately linked today. That's better cost structure, better availability, lower latency. Scaling down to the data center level, we run all custom Amazon design gear, all custom Amazon design protocol stacks. And why is that important? It's because the cost of networking is actually climbing relative to the rest of the compute. And so we need to do that in order to get costs under control and actually continue to be able to drop costs. Second thing is customers need more networking bandwidth per compute right now. It's East-West is the big focus of the industry. Because more bandwidth is required, we need to invest more fast. That's why we're doing private gear. Yeah, I mean it's a fascinating statistic because it's not just bandwidth. Why you said you do have up to 25 terabits per second between nodes, it's latency and jitter that are hugely important, especially when you go into databases. Can you talk about just architecturally what you do with availability zones versus if I'm going to a Google or a Microsoft, what does differentiate you? It is a little bit different. The parts that are the same are every big enterprise that needs highly available applications is going to run those applications across multiple data centers. That's so, the way our system works is you choose the region to get close to your users or to get close to your customers or to be within a jurisdictional boundary. From down below the region, normally what's in a region is a data center and customers usually are replicating between two regions. What's different in the Amazon solution is we have availability zones within region. Each availability zone is actually at least one data center. Because we have multiple data centers inside the same region, it enables customers to do real-time synchronous replication between those data centers. And so if they choose to, they can run multi-region replication just like most high-end applications do today. Or they can run within an AZ synchronous replication to multiple data centers. The advantage of that is it takes less administrative complexity when it comes. If there's a failure, you never lose a transaction where in multi-region replication, it has to be asynchronous because of the speed of light. Also there's a jurisdictional benefits too, right? Say Germany, for instance, with a new data center. Yeah, many customers want to keep their data in region and so that's another reason why you don't necessarily want to replicate it out in order to get that level of redundancy. You want to have multiple data centers in region, 100% correct. So how much is it that you drive your entire stack yourself that allows you to do this? I think about replication solutions. You used SRDF as an example. I've worked with that. I've worked for EMC for 10 years and just doing a two-site replication is challenging. Doing a multi-site is differently. You guys, six data centers and availability zone are merging. You fundamentally have a different way of handling replication. We do. The strategy inside Amazon is to say multi-region replication is great but because of the latency between regions they're a long way apart and the reality of speed of light, you can't run synchronous. If data centers are relatively close together in the same region, the replication can be done synchronously and what that means is if there's a failure anywhere, you lose no transactions. Yeah, so there's a great line you had in your session yesterday that networking has been anti-mores law when it comes to pricing. Amazon is such a big player. Everybody watches what you do. You buy from the ODMs. You're changing the supply chain. What's your vision as to where networking needs to go from a supply chain and equipment standpoint? Networking needs to be the same place where servers went 20 years ago and that is it needs to be on a more law curve where as we get more and more transistors on a chip we should get lower and lower costs in a server. We should get lower and lower costs in a network. Today in ASIC, which is the core of a router is always around the same price. Each generation we add more ports to that and so effectively we've got a more law in price improvement happening where that ASIC stays the same price and we just keep adding ports. So I got to jump in and ask you about open compute. Last year you said it's good. I guess I'm a fan but we do our own thing. Still the case? Yeah, absolutely. You're doing your own thing and just watching open compute which is like a make or fair for geeks. Open compute's very cool. The thing is what's happening in our industry right now is hyper-specialization. Instead of buying general purpose hardware that's good for a large number of customers we're buying hardware that's targeted to a specific workload, a specific service and so we're not, I love what happens with open compute because you can learn from it. It's really good stuff but it's not what we use. We want to target our workloads precisely. That was actually the title of the article I wrote from everything I've learned from you last year was hyper-specialization is your secret sauce. So you also said earlier this week that we should watch the mobile suppliers and that's where servers should be in the future but I heard somebody sent me a quote from you that said that unfortunately arm is not moving quite fast enough to keep up with where Intel's going. Where do you see, I know you're a fan of some of the chip manufacturers, where's that moving? Yeah, when I met with watch, arm and understand where servers are going, sorry not arm, watch, mobile and understand where servers is going is power became important to mobile, power becomes important to the servers. Most functionality is being pulled up on chip on mobile, same things happening in server land and so. You're saying it's mobile's a predictor of the trends in the data center. Exactly, exactly right. Because of the challenges with the form factor. It sounds like the form factor the importance of power and the importance of, well, density is important as well. So it turns out the mobile tends to be a few years ahead but all the same kinds of innovations that show up there we ended up finding them a service a few years later. All right, so James we've been at Wikibon to have a strong background in the storage world and David Floyer RCTO said one of the biggest challenges you had with databases is they were designed to respond to disk and therefore there were certain kind of locking mechanisms in place. Can you talk a little bit about what you've done at Amazon with Aurora and why you're fundamentally changing the underlying storage for that? Yeah, Aurora is applying modern database technology to the new world and the new world is SSDs at the base and multiple availability zones available and so if you look closely at Aurora you'll see that the storage engine is actually spread over multiple availability zones it was mentioned in the keynote it's a log structured store. Log structured stores work very, very nicely on SSDs they're not wonderful choices on spinning magnetic media so what we're optimized for is SSDs and we're not running it on spinning disk at all. So I got to ask you about the questions we're seeing in the crowd so you guys are obviously doing great on the scale side you get the availability zones which makes a lot of sense and certainly the Germany announcement with the whole Ireland, EU data governance thing and also expansion is great but the government is moving faster than some enterprises and so we were talking about that last night but people out there are saying that's great and it's a private cloud the government's implementing a private cloud so you agree that's a private cloud or is that a public? It's not a private cloud if you see Amazon involved it's not a private cloud our view of what we're good at and the advantages the cloud brings to market are we run a very large fleet of servers in every region we provide a standard set of services in all those regions it's completely different than packaged software what the CIA has is another AWS region it happens to be on their site but it is just another AWS region and that's the way they want it. Well people are going to start using that against you guys start parsing well if it's private it's only them then it's private but there's some technicalities you're clarifying that. It's definitely not a private cloud the reason why we're not going to get involved with doing private clouds is product software is different it's inefficient when you deliver to thousands of customers you can't make some of the optimizations that we make because we run the same thing everywhere we actually have a much more reliable product we're innovating more quickly we just think it's a different world. So James you've talked a lot that scale fundamentally changes the way you architect and build things. Amazon's now got over a million customers and it's got so many services just adding more and more Wikibon actually Dave Vellante wrote a post yesterday said that we're trying to fundamentally change the economic model for enterprise IT so that services are now like software when Microsoft would print an extra disk it didn't cost anything. For when you're building your environment is there more strain on your environment for adding that next thousand customers or that next big service or just do you have the substrate built that's going to be able to help and grow for the future? It's a good question it varies on the service and usually what happens is we get better year over year over year and what we find is once you get a service to scale like S3 is definitely at scale then growth I won't say it's easy but it's easier to predict because you're already on a large base and we already know how to do it fairly well other services require a lot more thought on how to grow it and end up being a lot more difficult. So I had some more questions before I go into some of the personal questions I want to ask you so I'm looking at these booths right here it's Netflix guys right there love that service awesome founder just what they do is this great company and I know they're a big customer but you mentioned network so at the Google conference we went to Google's got some chops they have the developer community rocking and rolling and then it's pretty obvious what they're doing and that's trying to compete with Amazon because it's too much work but they're going after the front end developer rails what not PHP and really nailing the backend transport you see in the peering really going after to enable on Netflix this next generation companies to have the backbone and not be relying on third party networks so I got to ask you so as someone who's a tinkerer mechanic if you will of the large scale stuff you got to get rid of that middle man on the network what's your plans you going to do peering Google's Aussie telegraphing they're coming down that road do you guys meet their objective same product better what's your strategy yeah it's a great question the reason why we're running private links between our regions is the same reason that Google is it's lower cost that's good it's much much lower latency that's really good and it's a lot less jitter and that's extremely important and so it's private links peering, customers direct connecting that's all the reality of a modern cloud and you see that do you have to build that in almost like you want to build your own chips I'd imagine on the mobile side with the phone we can see that everyone's building their own chips you got to have your own network stuff yep is that where you guys see the most improvement on the network side getting down to the down to that precise hyper specialized we're not doing our own chips today and we don't in the networking world and we don't see that as being a requirement what we do see as a requirement is we're buying our own A6 we're doing our own designs we're building our own protocol stack that's delivering great value and that is what's deployed private network is deployed in all of our data centers now yeah I mean James I wonder you must look at Google they do have an impressive network they've got the undersea cables is there anything that you look at them saying we need to move forward and catch up to them on certain pieces of the network I don't think so I think when you look at any of the big providers they're all mature enough that they're doing at that level I think what we do has to be kind of similar if private links are a better solution then we're all going to do it it makes a lot of sense does it you know deep packet inspection you know throttling traffic that just creates uncertainty so that I'm a big fan obviously that direction alright now a personal question so I talked to your wife last night and get to know you over the years here and Stu's obviously a big fan there's a huge new generation of engineers coming into the market open compute I bring that up because it's such a great initiative you guys obviously have your own business reasons to do your own stuff I get that but there's a whole new culture of engineering coming out a new homebrew computer club is out there forming right now my young son makes his own machines assembling stuff so you're an inspiration to that whole group so I would like you to share just some commentary to this new generation what to do how to approach things how you what you've learned how do you come up on top of failure how do you resolve that how do you always grow so share some personal perspective yeah it's interesting I know you're humble but you know interesting question I think being curious is the most important thing possible if anybody ever gets an opportunity to meet somebody that's atop of any business a heart surgeon a jet engine designer an auto mechanic anyone atop of their business is always worth is always worth meeting cause you can always learn from them one of the cool things that I find with my job is because it spans so many different areas it's amazing how often I'll pick up a tidbit one day talking to an expert sailor and the next day be able to apply that tidbit or that idea solving problems in the cloud so just don't look for your narrow focus your advice is talk to people who are pros in whatever their field is and it's always another James a friend of mine stay curious Steve Toddy actually called that Venn diagram innovation where you need to find all of those different pieces cause you're never going to know where you find the next idea so for the networking guys you know there's huge army of CCIEs out there some have predicted that if you have the title administrator in your name that you know you might be out of a job in five years what do you recommend what should they be training on what should they be working for to move forward to this new world the history of computing is one of the level of abstraction going up no never have the has it been the case those jobs go away the only people that the only time jobs have ever gone away is when someone stated a level of abstraction that just wasn't really where the focus is we need people taking care of systems as the abstraction level goes up there's still complexity and so my recommendation is keep learning just keep learning all right so I got to ask you the big picture now ecosystems out here Oracle, IBM these big incumbents are looking at Amazon scratching their heads and it's hard for us to change our business to compete obviously you guys are pretty clear in your positioning what's next outside of the current situation what do you look at that needs to be built out besides the network that you see coming around the corner and you don't have to reveal any secrets just philosophically what's your vision there I think our strategy is maybe a little definitely a little bit different from some of the existing old school providers one is everyone's kind of used to Amazon passes on value to customers we tend to be always hunting and innovating and trying to lower costs and passing on the value to customers that's one thing second one is choice I personally choose to run my SQL because I like the product I think it's very good value some of our customers want to run Oracle some of our customers want to run my SQL and we're absolutely fine doing that some people want to run SQL server and so the things that kind of differentiate us is enterprise software hasn't dropped prices ever and that's just the way we work enterprise software is not about choice we're all about choice and so I think those are the two big differences and I think those ones might last yeah that's a good way to look at that now back to the IT guy let's talk about the CIO scratch his head saying okay I got the facilities budget and it's kind of the I talked to one CIO he says I spend more times in planning meetings around facilities power and cooling than anything else on innovation so they have challenges here so what's your advice as someone who's been through a lot of engineering a lot of large scale to that team of people on power and cooling to really kind of go the next level and besides just saying okay throw some pods out there or what not what should they be doing what's their roadmap you mean the roadmap for doing a better job of running their facilities yeah there's always pressure for density and power is a sacred resource right now I mean power is everything power is the new oil so power is driving everything so they have to optimize for that but you can generate more power sure, sure, sure in space so they want smaller spaces and more efficiency the biggest gains that are happening right now and the biggest innovations that have been happening over the last five years in data centers is mostly around mechanical systems and driving down the cost of cooling and so that's one hot area second one is if you look closely at servers you'll see that as density goes up the complexity and difficulty of cooling them goes up and so getting designs that are optimized for running at higher temperatures and certified for higher temperatures is another good step and we do both so James there's such a diverse ecosystem here I wonder if you've had a chance to look around anything cool outside of what Amazon's doing whether it's a partner or some startup or some interesting idea that's caught your attention at the show yeah, yeah in fact I was meeting with Western with, pardon me, Hitachi data systems about three days ago and they were describing some work that was done by cycle computing and several hundred thousand cores we've had cycle computing going on oh wow last year we had a fun so Hitachi's just showing me some of what they gained from this work and then he showed me this bill and it was five thousand six hundred and seven dollars for running this phenomenally big multi-hundred thousand core project blew me away, I think that's phenomenal just phenomenal work James I really appreciate you coming in Stu and I really glad you took the time to spend with our audience and come on theCUBE again great pleasurable conversation very knowledgeable stay curious and get those nuggets of information and keep us informed thanks for coming on theCUBE James Hamilton, distinguished engineer at Amazon doing some great work and again, the future's all about making it smaller, faster, cheaper and passing those costs you guys have a great strategy a lot of your fans are here customers and other engineers so thanks for spending the time this is theCUBE I'm John Furrier with Stu Miniman we'll be right back after this short break