Oracle commits to 'parallel everything' architecture for Exadata, Exalogic

Oracle's CEO Larry Ellison insisted upon the importance of the "parallel architecture" strategy for developing hardware and software together.
Written by Rachel King, Contributor

SAN FRANCISCO -- There is the notion that if you design the hardware and software in concert, then you can do a better job than when they are split up by different companies, according to Oracle CEO Larry Ellison.

The goal for Exadata and Exalogic is to produce these products with the highest performance with the lowest cost, Ellison told audience goers at the opening keynote of Oracle OpenWorld 2011 on Sunday evening. This year's event will be hosting 45,000 attendees, making it the largest Oracle OpenWorld expo yet.

See also: Oracle OpenWorld: Hardware, Fusion, big data talk, SaaS on tap

Specifically, the goal breaks down twofold: The solution has to be faster than IBM's fastest computer, the P795, and it has to be more budget-friendly than X86 servers, world's cheapest servers.

Ellison defined the strategy as Exadata and Exalogic parallel architecture, a "fundamental architecture that says parallel everything."

From there, that solution breaks down to parallel hardware (servers, network and storage), and then parallel software (VM, operating system, database and middleware).

"When we first bought Sun, a lot of people said we're getting out of the hardware business," Ellison remarked, "I guess they didn't get the memo."

Ellison used the example of Apple, noting that the Mac computer maker is "doing a pretty good job of doing hardware and software, and online services that work together." As that works so well for consumers, Ellison posited, that should be translated to enterprise. He argued that this gives Oracle the opportunity to deliver better performance, security and reliability at a lower cost if all of these separate pieces are engineered to work together.

On the flip side, Ellison acknowledged that IBM is "a great company," but they don't have a fundamental parallel everything architecture that Oracle thinks is necessary.

"We're very good at moving data off of storage and on to the database server and back again," Ellison said. "It's all about moving data -- not about doing arithmetic in a microprocessor."

Ellison argued that no one else in the world is tapping into this problem except for Oracle.

"Another piece of magic," Ellison described, is hybrid columnar compression, which equates to 10 times the amount of data compression. The data movement is also up to 100 times faster, and there is 10 times less data to move.

"When we announced Exadata a few years ago, we had numbers that no one believed," Ellison said.

To sum up, the new Exadata database machines rack up the following stats:

  • 10 to 50 times faster queries
  • Four to 10 times faster OLTP
  • Up to 10 times less disk storage
  • Fault tolerant (no single point of failure)

Compression has another interesting derivative, Ellison said, explaining that the new three-tier storage hierarchy requires a lot less disk, but a lot more memory, DRAM and flash. As far as the cost side goes, Oracle is promising lower cost components and compression, with commodity PC storage (i.e. DRAM, Flash and disk).

Ellison admitted that "hardware always breaks," adding that "software breaks too," but he argued that if systems have a parallel architecture, you should be tolerant of those failures. Systems might be slower, but the but the overall application should keep running because the architecture is inherently parallel.

So far since Exadata was introduced, there have been 1,000 units installed. The goal is to see another 3,000 installed this year. Customers include Starbucks, Sears, Virgin Mobile, Samsung, GE, and Procter and Gamble, among many other worldwide.

Companion to that is Exologic, which runs Oracle middleware. The goals here specifically point toward 10 times faster Java applications with up to five times more users. But on Oracle EBS and Siebel applications, customers will see only four times faster response times and three times more users.

The parallel everything architecture will also be extended to the SPARC SuperCluster machine, that runs Solaris and the same fundamental architecture. A new generation of SPARC Solaris T4 servers will be up to five times faster than the T3, as fast as Exadata for Database and as fast as Exologic for Java.

Ellison also introduced the Exalytics instantaneous business intelligence machine, which is fully parallel and designed for in-memory analytics (i.e. prediction, simulations, scorecards, etc.) and delivering results at the "speed of thought." Essentially, Ellison said, "as fast as you can type."

All the data being analyzed is in the main memory -- 1TB of DRAM on the main box -- with 40 cores of CPU cores with Intel Xeon processors. The entire hardware scan rate is roughly 200GB per second, so a 5TB database can be scanned in five seconds.

As far as latency goes, Ellison affirmed that "there is none."

Ellison asserted, "We're determined to deliver best in breed in every aspect in our computer architecture."

Editorial standards