 Hi, welcome. Jeff Frick here in the Palo Alto offices of Silicon Angles theCUBE. We're here for a CUBE conversation with David Floyer, CTO and co-founder of Wikibon. David has been following data centers and infrastructure for quite some time. He's really excited about some of the impacts that Flash is having and some new technologies to really change the game, if you will, in the data center. And clearly there's some big trends happening with the hyperskill companies really driving ahead for the enterprise. So we've had some big announcements, David, in the last week or so, and you've got some new research out that we want to talk about. So why don't we jump in? What's happening in the Flash world that's exciting? Well, at Pekona 2014, the conference, there was the announcement of the NVM non-volatile memory extensions to MySQL. And that's a pretty big deal because for the first time, the ability to do some of the fancy things with Flash, making applications Flash aware, has gone from the universities and the people who are the leading edge and to a standard NVM extensions for all of the MySQL distributions, both for Oracle, for MariaDB, and for Pekona. All of them now have these NVM extensions. So let's jump in a little bit. What does that mean exactly, the NVM extension, and why is that important? So there are two major components of the NVM extensions. The one that's been around for some time, certainly from a testing point of view, which is atomic rights. So when you're writing databases, if you are using a standard MySQL, you're going to write twice. You're going to do what's called double rights. And that allows things to be firmly guaranteed that they will have been written and that you can recover from them fundamentally. So what atomic rights do is they allow you, instead of having to write twice, to write once. By doing so, you can effectively reduce the number of rights by 50%. And you get about a 70% throughput advantage from those. So it's a pretty big impact. Pretty big impact. The second thing that's just been added, which is totally new, is the compression, NVM compression. Now in MySQL, there's been a compression around for a long time, but it's a big overhead. It's about, if you've got it, it can take 100 transactions, per second, it'll go down to 50. The overhead is really very significant indeed. So people have often not used that. But if you now can use it, a pretty well no overhead, it's zero overhead for using the compression. And in addition to that, it uses flash sensibly. It's treating it as memory in one aspect, it's writing to it very fast. But with the other way, it is knowing that it is flash. It is optimizing the way that it writes to that flash. So you get, by writing only half the number to compression, you're going to double the life of the flash. And that's a big thing for flash. And by writing sensibly, you've got another doubling as well. So it's four times as efficient on the flash drive itself, on the flash PCIe card itself, by using the compression. Those are two big impacts of it. And it's gone from being just a drive, being as a thumb drive, for example, in your PC or a disk drive, just replaced with an SSD with flash. It's gone from that now to being something that contributes to the throughput through the system very significantly indeed. Right. And so this is part of the kind of flash aware applications that people are talking about where you use flash, not just a way to write faster storage, but actually an opportunity to change the way you design your database, change the way you design your application to really take advantage of the technology that's so much better than spinning rusty disks, as everybody likes to say. Absolutely. And better than just copying the protocol of those spinning rusty disks, which were 50 years old and pretty slow and pretty inefficient. So yes, making it such that you can use direct PCIe bus APIs and extensions to reduce that overhead as much as possible and use it to extend the life of the flash itself. It's a pretty big deal. So in fact, if you install the NVM extensions, you get a 10% because it's done some work behind the scenes just to improve things as they were. So you get about a 10% extension. And I think there's a chart which we can... Yeah, Dave has done some work on this. So as always, everything that the Wikibon group does, Dave, Jeff Kelly, Stu Miniman, is available online at wikibon.org. It's open source research. We encourage you to go on there and search. But we have grabbed, I think we've pulled up now, the chart from Dave's article, which is entitled just in case you're watching at home. MySQL receives over three X boost and performance from PCIe flash. So Dave, why don't you go ahead and take us through the chart? Yes. So the base of this is somebody who's running an application which has compression working and it's using double writes. It's making sure that everything meets the atomic requirements of a true database. So that's the base, which is 100. Now, step one is that you can put in these NVM extensions. And as I said, just by using Flash with these, you'll get some 10% improvement. They've just been speeded up and tied it up from 50 years of garbage. Step two is to add the atomic writes. And that goes from 110% to 170% of the base of 100. So now we are allowing 70% more throughput through the system. Step three is we're running compression and the overhead of compression is about 50%. So we can double the throughput now and go from 170 to 340% improvement. So by putting these NVM extensions in their open source by using PCIe card, you can't use these extensions on SSDs because that's still using the old protocol. But by running it on a flash card that will use the protocol, for example from Fusion IO, will be able to use this protocol, you can now get three times the throughput of the application that it had before. Wow, that's pretty significant. So who are the people that are going to benefit the most from this? Well, the people that are going to benefit the most are people with large SQL applications. They're often the cloud providers. They haven't had the resources and they don't want the overheads of investing in very large Oracle or SQL Server or DB2 type of databases. So they've been using MySQL, it's become the de facto standard for open source, facto standard for large scale web-based databases. So it's been used very, very extensively. There's obviously also a lot of other databases out there in the open world, the non-SQL databases for example, but there's a lot of advantages of running SQL. It's sound, it's less complex, much less complex for the developers than running no SQL. So the speed to change, speed to get things up is faster with MySQL. So it's a sound database and it's people who are running large-scale applications that are really going to benefit from these enhancements. Every time you have an investment in a database, especially in a fast growing area, you will need to re-architect it, you will need to shard it maybe, you'll need to do things to extend that database. If you can get three times the throughput, you've essentially enabled three times the life basically of that implementation and that is a big, big deal, big, big deal. And as you said, this initial instance is really around the MySQL database in that community because of the open source and the innovation. But you said it will be coming in the not too distant future to some of the other databases. It's going to be a business as normal way of doing things. It's going to be embedded in Oracle. Obviously, it takes much longer to get into the design and to the testing that they do and much more mission-critical applications like that. But it will get there into my Oracle, it'll get into SQL Server, it'll get into DB2, it'll also get into the non-SQL databases as well. The key pair components already have parts of NVM extensions which will help them as well. So it'll become ubiquitous over time. And Flash is still the number one NVM. There's a lot of talk about other things maybe coming in. But Flash is it for the moment. Maybe in five years time we'll see something else. But Flash is going to be around for a long time and it's driven by the consumer use of it. You and I sitting here without the old-fashioned disk drives. We have Flash-only types of PCs. And again the other interesting thing here is there's benefits that can be recognized immediately, correct? Correct. By putting this in just in terms of the speed. And then there's longer-term benefits by people really re-architecting their applications and databases around this capability. I mean that's the most exciting thing of all. We've had such a built-in restriction on the number of database calls you can make in a transaction that's been built into every application that's been built for the last 50 years essentially. And this allows 10 times the number of database calls per transaction to be put through. That's a huge difference. 10 times 100 times. A huge difference in how you design applications and how you can make them usable by everybody. And all businesses are trying to go to reduce the number of, for example, in SAP. They're trying to reduce the number of landscapes that they have in SAP because they want the same database to be supporting every division within an organization. They really couldn't get there before. So with techniques like this, they can put in one landscape for the whole of the division. And we were looking in detail about what that meant to those organizations. They could get rid of tens, tens of financial analysts in each of their divisions having to do manual work to reconcile the financial books at the end of each month or end of each quarter. So major advantages in improving the productivity of organizations by reducing the number of databases. And then the other thing that does come up time and time again at some of these shows, in talking about flash, is this concerned about the flash longevity and lifespan and replacement and these types of things. And you're saying there's even this benefit here where by being optimized or right because it is flash, you can take some of that negativity away from flash, if you will. Absolutely. The alternative way of using this is to take an application which has not got compression on it and to add compression on it. And by adding compression, you get half the number of writes and therefore you've doubled the longevity. But in addition to that, because it's doing that NVM compression in a smart way, using its knowledge of how the best way of using it, using some pretty good techniques, you can again have again the number of the longevity of that flash. So if you do that, you extend the life of the flash by a factor of four for free. You're not no overhead on it. So it's absolutely no brainer of a decision to use this as a way of speeding things up and as a way of improving, just improving the application. All we're missing is Gordon Moore, right? We have better performance, better performance, more longevity, no extra overhead and the continued driving pace of innovation just keeps going. Yeah, it's exciting. It's exciting to see it come to a stage where it's really not just a drive but now it can be using the characteristics of it as memory to really go towards large-scale memory systems, which have this fantastic additional ability to be persistent instead of losing it every time you have a hiccup in the system. And latency too, right? Which everyone is very impatient as we walk around with our phones and we want to answer every question in just a few milliseconds. If not yesterday. If not yesterday. Well, thanks for stopping by and kind of breaking that down for us. That's what we like to do. We like to, you know, we look at the news. We like to dig into a little bit deeper. We get the smartest guys in the room. Sometimes they're our own, like David here, to really drive down. Again, I want to point you to his post on this again at wikibond.org. It's MySQL receives over 3x boost in performance from PCIe Flash. So take a look at the article. Feel free to contribute. Everything on wikibond.org is open source, as are all the CUBE interviews and CUBE conversations. So thanks for joining us, Dave. Thanks for stopping by and breaking it down for us. That's all for today from Palo Alto. We'll see you next time. Thanks again, Jeff Frick from Silicon Angles theCUBE, signing off.