Oracle CTO Ellison To Cisco, EMC: Bring It On
Ellison In His Element
Larry Ellison took the stage in the conference center of his company's headquarters Jan. 21 to unveil Oracle's X5 line of converged systems to staff, partners and customers.
Clearly comfortable in his new role as CTO, Ellison's passion for engineering was evident as he talked about the technical capabilities of the Virtual Compute Appliance, Exadata Database Machine and other upcoming Oracle products.
While he may have been wearing his engineer cap, Ellison was still ever the salesman, ever the combatant. With typical bravado, Oracle's founder threw down the gauntlet to Cisco and EMC, proclaiming he was ready to take them on in their own game of selling low-priced servers to power modern data centers.
Here's Ellison in his own words.
Gen 5
We're here to introduce the fifth generation of our engineered systems. Again, the idea is taking the hardware and software and during the engineering phase actually designing them to work together. We do the integration so you don't have to. But more than that we locate functions at the right place in the stack. If it's a storage function, we locate it closer to the storage. If it's a compute function, it's close to the compute service, and there's security through all layers, through the storage, through the networking and through compute.
This is our fifth generation of engineered systems called X5.
Data Center Trends
The biggest data center trends are these two-socket Intel servers which have a very, very low purchase price. People are building the core of their data center around these two-socket Intel servers. And almost always running Linux.
So it’s a lot of two-socket servers, pretty much the cheapest servers you can buy, running an open-source operating system and people are building that as the core of their data center, and it's very attractive because it has a low purchase price.
You can make the argument, we make the argument, that these do-it-yourself data centers are expensive. You do the integration rather than we do the integration.
Price War For Data Center Core
We've really never aimed at the lowest-possible purchase price with our engineered systems. We've never really competed for the data center core, those two-socket Intel servers. We focus on something very different. We said, OK we're going to do all the integration for you, and by doing that, by engineering every piece, the storage, the networking, the compute servers, and all the software on those three layers we can deliver extreme performance.
And our very first engineered system, Exadata, did just that. It delivered extreme performance, and we just have made it faster and faster and faster as we went from disk storage to disk storage fronted by flash to, you'll see with generation five, some options that are all flash.
Competitors And Imitators
We didn't introduce engineered systems; we didn't invent the idea. Teradata had a converged system, we call it an engineered system, the industry now calls it converged systems, the idea of taking storage and networking and compute and putting them together.
And then we came along with Exadata, and I think we did the better job. We certainly sold more systems than they did, and we still sell more systems than they do.
And then we came out with other engineered systems, Exalogic, Exalytics, more and more and more all aiming at the highest performance and the best cost-performance. There were even imitators. IBM came out with PureSystems, which was very interesting. Didn't do so well, but interesting. Copying you is the highest form of flattery.
Cisco And EMC
Then something very interesting happened in the industry in that Cisco Systems developed a partnership with EMC and they built, if you will, an engineered system, or a converged system, taking storage from EMC and software from EMC, particularly VMware, and Red Hat's operating system, and Cisco's switches and Cisco then bought some two-socket servers and put them together.
But the thing that Cisco did, and they have been very successful, is that they went and decided to compete for that data center core with a very low purchase price. Interesting. I mean they aren't close to us in performance, but they delivered a very low purchase price.
Original Converged Strategy
Our strategy for Engineered Systems in the very beginning was to deliver extreme performance, the highest performance, and once we did that we would have better cost-performance than anybody.
You take 100 applications, put it on Exadata, and that's better than running those 100 applications anyplace else in terms of your overall cost of ownership.
But the purchase price of an engineered system was still an impediment for some people buying Exadata. They believed us on the performance, but is it really true, I mean it's an expensive computer, it costs hundreds of thousands of dollars? A really big one can cost more than a million dollars. How can that be the cheapest one?
Well it is. But we're tired of having this argument.
Shifting Strategy
So we have a new strategy. We're going to compete for that two-socket data center core business.
And from what I can tell, the way you compete, because it's two-socket servers running Linux, is you just have the lowest price in the industry. So our strategy, and you'll see it here, is to deliver not only the highest-performance appliances and machines in the industry by a large margin, but now, with the fifth version of our engineered systems, to deliver by far the lowest price for the data center core, and that's what we're going to do.
We didn't invent that idea, I think Cisco and EMC did, but just like Teradata invented the database machine, we think we can do a better job.
Virtual Compute Appliance
The Oracle Virtual Compute Appliance, VCA, is converged infrastructure, with compute servers, software-defined networking and storage, all engineered to work with software, all in one box.
The thing that’s so attractive to purchasers when they're building up the core of their data center is the lowest purchase price in the industry by far. And everything else is gravy.
So we'll tell you that we can deploy applications in a couple of minutes because we have application templates, that's part of our software stack, and other people take hours and days to deploy applications. But that doesn't have to be how you make your decision because we've got the lowest purchase price in the industry.
Price Comparison
If you look at this VCA box it includes 27 two-socket servers, latest Intel servers, that's almost a thousand cores.
It includes the rack, includes everything, power supplies. It's the whole megillah, as my mom would say. And it's around half the price of Cisco, looking at just the hardware component. This is not including software. Well, let's take a look at the software that goes with UCS. So you have Red Hat server, which has no purchase price but has an annual support fee, and you've got VMware, which does have a purchase price and a support fee, and then you have some management software.
So the hardware, much lower price, the software much lower price.
VCA Storage And Total Cost Comparison
VCA comes with different storage options. It has a Fibre Channel interconnect so you can connect with your existing storage if you want to, but we also will bundle in storage and you have some options. You can have network-attached storage, or you can have a storage area network.
So if you look at these totals, it costs, for VCA plus NAS, plus all the networking, plus all the software, --by the way this is 27 two-socket servers, this is a lot of compute power, a lot of storage, powerful network interconnect -- is about $60,000, plus $70,000 support. Compared to more than double the price, with almost triple the support costs.
Aggressive Pricing
We are going to aggressively compete with version five of our engineered systems. We are going to aggressively compete for that two-socket core of your data center, and the way you do that is you get aggressive with pricing and we think we've done that.
This is a new strategy for Oracle using Virtual Compute Appliance integrated with software-defined networking, and a couple of storage options to go after the data center core.
FS1 For Storage Area Network
Our old strategy was: highest performance, best cost-performance. Our new strategy is: highest performance, lowest list-price, best cost-performance. We're just trying to make the decisions easier.
Here is FS1 compared with EMC. We're not cheating, we're not picking out an expensive EMC box to compete with our low-end FS1. This is our fastest, most-expensive FS1 you can buy. We don't make a flash system more expensive than the FS1. This is as much as you can spend, this is as fast as you can go. It's much faster than EMC's fastest flash system. And you can see, apples-to-apples comparison, comparing the same amount of capacity, we are less than one-third of their purchase price.
By the way, it runs much faster. But that's gravy.
FS1 Attributes
Fully fault tolerant. Petabytes of flash memory. And if you balance reads and writes, and everybody knows that writes are more expensive than reads, but if you do 50 percent reads and 50 percent writes, it's still a couple of million I/Os per second.
You can cut your capital expenditures by half by combining up Virtual Compute Appliance and FS1 and you can deploy your applications in minutes, save a lot of time and a lot of labor, and it runs anything in your data center.
A lot of people think our appliances are designed to just run the Oracle database or Oracle middleware or Oracle analytics. This runs anything. It runs your Microsoft applications. It runs anything in your data center.
Advanced Interconnection
Internally, this thing is interconnected by Infiniband. Now you never see Infiniband.
Infiniband is interesting because it's much faster than any other interconnect, but it's also much more reliable than any other interconnect and has guaranteed message delivery in a certain period of time, makes recovery much simpler.
The latest interconnect technology, in terms of the hardware interconnection, is Infiniband. But on top of that, you wire this thing once and then you use software. If you want to reconfigure your network, add more nodes, add more storage nodes, add more compute nodes, you reconfigure and expand by changing parameters in a file. The whole network is defined by software, so you wire it once and you're done. All that's for free.
One Vendor
If you look at what we're trying to do versus what Cisco has done very effectively, we're taking, we're engineering all of our components, so they work together. And we think the fact that we've engineered all of these pieces ourselves, one vendor, and we've designed them to work together, we've built them to work together, we support them together, we test all the software to work together with the hardware configuration, and then when we issue patches, all that stuff's been tested.
That’s very hard to do when you have multiple hardware vendors and multiple software vendors.
They're all issuing different patch releases at different times. The one responsible for making sure all those patches work together would be you.
Stacking Up To Cisco
We're about half the price of right now of the most popular converged system in the data center core, which is Cisco UCS, and we're very interested in pursuing that business.
And again, when you pair up VCA with FS1, you pay half as much, but you have to be willing to go twice as fast.
Oracle Database Appliance X5
We've sold thousands of them. Not millions. We'd like to sell millions. We've only sold thousands, to a variety of industries.
ISVs love them, people who develop their own applications love them. They package their applications and ship them out on the Oracle Database Appliances.
Again, low cost. We've upgraded with latest Intel servers, added more flash capacity, software-defined networking and faster networking. But maybe the biggest thing we've done is just constantly improved our software.
We are constantly improving and relocating software to the right level of the stack to deliver much better reliability, much better performance and much easier to use.
A database appliance costs less than rolling your own using components from Dell.
Zero Data Loss Recovery Appliance X5
Not every appliance is about running faster, or even having the lowest purchase price, though this does. The big thing about the Zero Data Loss Recovery Appliance is that it doesn't lose data. I mean it sounds funny, well, of course, it doesn't lose data, none of my recovery systems lose data, that’s absurd. That’s got to be the case.
Really?
So the focus of what's called the Zero Data Loss Recovery Appliance was to completely automate logging and recovery, take human error out of it thereby having low cost, and never lose any of your data. That’s a differentiator.
This is about as no-brainer an appliance as we have.
Big Data Appliance X5
This is our appliance that runs Hadoop and our Oracle NoSQL database, but we also have a layer of software called Big SQL that allows you to take data out of the Oracle database and join it or combine it with data from a Hadoop database and a NoSQL database. You can integrate, you can look at all your data, whether it's stored in a NoSQL database, a SQL relational database. By the way, NoSQL doesn't mean no SQL. It means not-only SQL.
You've got a comprehensive, 360-degree view of all of your data with Big Data SQL.
You get all this is you're willing to pay less, but only if you're willing to pay less.
Exalogic X5
We've had Exalogic for a while. It runs our middleware, it runs our applications, it runs any kind of Java application.
The interesting thing about this is it's designed as a private cloud. In other words, we run Exalogic, the software that runs in Exalogic is the same software that runs in the Oracle cloud. That's a very important thing.
Exadata Database Machine X5
Our very first engineered system ever was Exadata. This is the fifth generation of the Exadata machine, the fourth generation that we've built ourselves.
It's designed to run all Oracle database workloads, and there are lots of different database workloads. There are huge data warehouses, there are very fast real-time transaction processing systems. I mean very, very different kinds of workloads that the Oracle database has to support and this is designed to run all of them well.
It's the latest Intel processors, it's the latest Infiniband, coolest storage and most importantly the fifth generation of software we've developed for Exadata. It's fault-tolerant, there's no single point of failure in this system.
You start small and you scale out.
A Disk Argument
It's very interesting that Juan Loaiza [senior vice president, systems technologies, pictured] and I used to have these huge fights about high-performance disk versus high-capacity disks.
And I thought we should only have high-capacity disk because everything else would be in cache. And Juan says, doesn't matter, people want high-performance disk.
Anyway, Juan won the argument.
Argument Settled
Lo and behold, with the latest generation we've gotten rid of our high-performance disks. I'm not sure I can claim this as a victory though, because we've replaced it with flash disks basically for the same price. It's really astonishing.
We've thrown out our high-performance disks, we haven't raised prices at all, and give you about the same amount of flash as we used to give you for high-performance disk at the same price. As those prices came together, then finally high-performance disk became irrelevant.
High-performance disk doesn't look very high performance when compared with extreme flash disk. Much higher bandwidth, much higher I/O rates, dramatically faster system with fifth generation of Exadata.
Down To Silicon
In another presentation sometime in the future I will talk to moving stuff out of software and into silicon. We're doing a lot of it.
Always pushing things down closer and closer to the silicon. You can only do it if you work on the compute layer, work on the networking layer, work on the storage layer and work on the silicon layer. And we happen to do all of those things, which gives us a huge advantage in being able to deliver extreme performance, which we've done for years. And now we're using that engineering not to deliver extreme performance, but what you can call an extreme list price.
How Fast Is Exadata?
There are a number of features we've added to this system to make it much faster. Not simply a hardware upgrade, but a hardware upgrade in many levels -- storage, networking and compute -- and a software upgrade at many levels as well.
How fast is it? Really fast. Really fast. There's almost nothing to compare it to.
We can't find anything that's within a factor of 10 as fast for analytics.
We're giving you for the first time with Exadata elastic configurations. So you really can invent your own configurations.
SuperCluster
Maybe you know this, maybe you don’t: Our SuperClusters, and we have two different models of SuperCluster, use the same storage server, and the same storage server software, as Exadata.
And all of the upgrades we've made to Exadata also go into both of our SuperClusters. Including Extreme Flash I/O, a lot of advanced software optimizations are in both places.
Data Center Of The Future
The next big trend in computing, everyone's talking about the cloud, but people are not doing a terrific job of connecting. Data centers are not going to go away, by the way. Clouds are going to get bigger, clouds are going to get more popular and data centers are not going to go away. And one of the things we're going to be focusing on is how we interconnect our data centers to the cloud.
You have to have the ability to press a button and move an application from the cloud to the data center and back again easily and gracefully. You have to have the ability to take a database, take data, and move it from your data center into the cloud and back again.
Cloud Compatibility
So there has to be some degree of compatibility between the public cloud and your private data center and that's another important initiative for us, that the things we sell you to run in our data center gracefully connect to things we deploy in our cloud so you can move applications and data back and forth and back again on an as-needed basis.
And that really is the data center of the future, the combination of all three, the data center core made up of two-socket commodity servers surrounded by purpose-built engineered systems and appliances interconnected to the cloud.
That is the future. That is what we're working on. That's what we'll continue to work on.