Original URL: http://www.theregister.co.uk/2010/01/29/oracle_sun_exadata_ibm/

Ellison: Only Oracle can do OLTP clustering

Armonk, we have a problem

By Timothy Prickett Morgan

Posted in Servers, 29th January 2010 08:02 GMT

You have to admire Larry Ellison, the Oracle co-founder and chief executive officer of IT giant Oracle. Well, maybe admire isn't the right word. But you can at least be amused by him.

Yesterday, Ellison spent the last hour of the five-hour Sun takeover extravaganza taking questions from the audience, mostly fielding softball queries about the America's Cup boat race or why he should buy the Golden State Warriors basketball team. But mixed in among the sycophantic admiration for Ellison and Oracle, there were, thankfully, a few serious things said by Ellison and a handful of decent questions.

"We're excited to be in all of these new businesses with these fantastic people from Sun," Ellison started off saying, referring to the "fabulous Sparc chips" and to Solaris as the "best high-end Unix operating system in the world." Oracle intends to put those Sun hardware and software engineers to work, making integrated and highly tuned application, database, and middleware machines. The kind of thing that the IBM of the 1960s put together with a silly old thing called the System/360 mainframe.

"It's not that it hasn't been done before," Ellison said. "That strategy made IBM the most important company in the history of Earth. So we kind of like that strategy. So we are going to adopt it."

But don't get the wrong idea. Oracle is going to sell all of the chunks of its software stack to run on other hardware and software platforms as well. Oracle can have its cake and eat yours too. So if customers want to put Oracle's database on Hewlett-Packard servers running Windows, that's cool, or maybe they might want to put middleware on Dell boxes. Well, go for it. "We're in both businesses: the best-of-breed component business and the integrated systems business."

IBM decided it could make more money getting out of integrated systems, which it started to do in the 1980s. It sold components and then, later in the 1990s, it sold services to integrate and manage systems that customers cobbled together. IBM has made a fortune on this bet, and it did so because of the onslaught of much cheaper systems - first minicomputers, then Unix boxes, then Wintel machines - that could do the work of IBM's proprietary midrange and mainframe boxes.

In days gone by, IBM was a dominant supplier of software on its System/3X and AS/400 minicomputers, but it gradually sold off those businesses so it could become the preferred and presumably neutral supplier of systems on which all kinds of applications run. IBM just wanted its share of the server and storage sales and more than its share of the services revenues, which added together gave the company immense account control at some 500,000 businesses in the world. It sure beat going bankrupt, which was Big Blue's other option in the early 1990s.

Because there is only currently one co-branded Sun-Oracle product, the Exadata V2 database appliance, that was what Ellison and a lot of his colleagues talked about. That cluster, which was rechristened the Sun|Oracle Database Machine at yesterday's event - although Ellison didn't appear to get that memo - was announced last September, many months after Oracle had hoped to close the Sun deal.

The Exadata box combines Sun's X4170 rack servers (used for database processing), X4275 rack servers (equipped with disk and flash drives and running Oracle's Exadata storage software, which has optimizations to run data warehousing and online transaction processing (OLTP) queries), and Sun's InfiniBand switches to glue them together. The database servers are equipped with Oracle 11g Release 2 (presumably with the Real Application Clusters extensions for parallel database processing) and Oracle Enterprise Linux, and the storage servers run Exadata Storage Server Release 11.2.

So, how's it doing?

Up until the September launch of the Exadata V2 database appliances, Oracle had been peddling a prior generation of gear based on HP iron, which was summarily pushed off a cliff. Up until that point, Oracle had reportedly only sold 25 of the HP|Oracle Database Machines, and it had only been aiming them at data warehousing workloads. But with the addition of flash memory and the optimizations that Oracle put in the database and storage software with Exadata V2, now the box can do random OLTP processing as well as the more sequential data warehouse dicing and slicing.

So how its Exadata V2 doing? Better than expected, apparently. Ellison said that Softbank - one of the largest telecom providers in Japan, who just so happens to be Teradata's biggest customer in the Asia/Pacific region - has picked up an Exadata V2 setup that can do some database crunching work in 4 hours that used to take 30 hours on the Teradata machine. Because of such performance improvements, Ellison said that the pipeline for Exadata V2 machines in North America alone were "measured in the hundreds of millions of dollars." (Ellison was not more precise, unfortunately).

John Fowler, the former Sun exec who is now executive vice president of hardware engineering at Oracle, said that interest in Exadata machines for proof of concepts was outstripping Sun's ability to deliver gear to prospective customers. He did not discuss numbers either.

Ellison was asked if he took a threat from IBM seriously, considering that Big Blue has similar hardware and database assets.

"They do not have the same assets," Ellison said with a bit of condescension. "The Oracle database scales out; IBM DB2 for Unix does not. Let me see. How many servers can IBM put together for an OLTP application on what used to be called DB2 UDB - IBM has two databases, one on mainframes, one on Unix, totally different products - let's see, how many of those can they group together? One." Laughter ensues.

"They can have one servers attacking really big jobs. When they need more capacity, they make the server bigger. That's all they can do for OLTP. So they don't have the same assets," Ellison admonished. "You would have thought, years ago, that IBM would have come up with its own Database Machine. I mean, it's so obvious. They've got the hardware, they've got DB2. It's fascinating.

"We think we have a huge advantage over IBM in the ability to scale out to cluster systems for OLTP. IBM can't do that, and that will be a big problem going forward." And wait for the punch line. "IBM doesn't have the same assets, and that is a big problem for them. They don't have Java, and they don't have the Oracle database. What they got is a problem."

Ellison said that he believed Oracle has a ten year lead on IBM, and it didn't bother to mention that IBM is, in fact, cooking up its DB2 PureScale OLTP clustering for DB2 running on Unix, which was previewed last October and which was supposed to start shipping at the end of last year. IBM has been quiet as a mouse about DB2 PureScale and clearly needs to get it ramped up and competing with Exadata V2.

As Ellison reminded the crowd, IBM does offer clustered DB2 on mainframes (which is based on Parallel Sysplex clustering that predates Oracle RAC and is arguably still better). He referred to it as a "good product" with only one fault: it only runs on mainframes, rather than on "modern computing systems" as Oracle's database does.

IBM has offered database clustering on its OS/400-i/OS platform, called DB2 Multisystem, since 1995, well in advance of Oracle RAC. While the DB2-Parallel Sysplex combo does give IBM plenty of revenues and profits, DB2 Multisystem has not been leveraged by midrange shops that are basically allergic to complexity and that much prefer SMP systems despite the extra price you pay for hardware and software.

But Oracle has been on the clusters kick for nearly a decade now, and as networking, memory, and other system components can be brought to bear, it may be very difficult to argue SMP systems are better if they are so expensive compared to clusters. IBM - and indeed any platform provider - needs to be able to sell clusters for OLTP and data warehousing alike. Oracle has that right, and if it didn't, IBM would not have rushed out the DB2 PureScale announcement so quickly last fall.

Bootnote

Oracle took the "Sun" part of the Sun Microsystems logo out of italics and dropped the Microsystems entirely. Larry don't sell no stinking Micro anything. But he is the Sun King now. ®