Oracle's Ellison Tries To Outmaneuver SAP Hana

There's a tactic in sailboat racing whereby a competitor who is trailing can catch up on a downwind leg by positioning to catch the wind behind the lead boat. This is precisely the position Oracle holds in the in-memory database race and precisely the tactic Oracle CEO Larry Ellison employed on Sunday night by announcing the Oracle 12c In-Memory Option during his kickoff keynote address at this week's Oracle OpenWorld conference.

By promising dramatic improvements in performance with zero disruption, Ellison is vowing to deliver much the same benefits SAP has been promising from its Hana database for more than three years. But for Oracle database customers -- and there are hundreds of thousands of them -- the prospect of gaining in-memory benefits while sticking with the product, the training and tooling they know is likely to be very compelling. By comparison, Hana is a relative unknown deployed by only a few hundred customers.

Never mind that the vast majority of Oracle customers will have to upgrade to release 12c before they can add the In-Memory option. Oracle customers have been through these release upgrades before, and Ellison is promising continuity.

"You flip a switch and all your existing applications run much faster," said Ellison to applause Sunday night. "There are no changes to SQL. There are no changes to your applications. There are no functions that are restricted. Everything that works today works with the In-Memory Option turned on and there's no data migration."

[ Want more on Oracle's latest financial results? Read Oracle Earnings Rebound: All Eyes On OpenWorld. ]

The promises went on. Analytic performance is improved by at least 100 times while transactional processing speeds are doubled, according to Ellison. The In-Memory Option works in both scale up (symmetric multiprocessor) and scale-out (RAC cluster) deployments, and "things that used to take an hour now take a few seconds," Ellison said.

The In-Memory Option is described as compatible with existing servers (either from Oracle or third-party vendors), and Ellison made a point of the fact that customers will have a choice of putting some tables in memory, others in flash and others on disk, depending on query priorities. Exadata makes this choice automatically based on query patterns, with the new In-Memory Option making DRAM a richer high-speed option. Hana, by contrast, runs all in DRAM, but that means even low-demand data must be managed in comparatively high priced DRAM storage.

Demos during the keynote showed a simple query improving from 2 billion rows per second using 12c to 7 billion rows per second using 12c with the In-Memory Option. This was on a generic Intel X86 two-socket server. Ellison also introduced a new version of Oracle's top-of-the-line M6-32 Sparc server dubbed the Oracle Big Memory Machine, with 32 terabytes of DRAM. A demo showed scan rates on this monster machine running at 341 billion rows per second.

Despite all the promises, a few key facts were missing from Ellison's keynote, starting with release data for the In-Memory Option. Ellison said the M6-32 server is available immediately, but there were no details specifically on the new database option. Microsoft is another database vendor playing catch up on in-memory capabilities, but its Microsoft SQL Server Hekaton release is already in community technology preview and is expected to see general release before the end of this year.

Without beta tests or production deployments, there's no way of knowing whether the performance promises will hold up. The In-Memory Option introduces a new dual storage approach whereby data is saved in both rows and columns. Ellison insisted the new design will improve both transactional and analytical performance, but SAP raised doubts.

"SAP Hana has one columnar store for both transactions and analytics whereas Oracle has two and is, hence, bloated," said Amit Sinha, SAP senior VP of database and technology, in a post-keynote email comment shared with InformationWeek. "This approach implies a 5x greater data footprint and more DBA-intensive work to designate which rows to change to in-memory."

The design will also introduce a drag on transactional performance due to the overhead involved in keeping OLTP and OLAP in sync, claimed Sinha, but Ellison promised quite the opposite -- a doubling of transactional performance because analytical indexes, and updates thereof, will be eliminated.

SAP CTO Vishal Sikka posted a video blog on Sunday in which he welcomed Oracle to the in-memory club. "The good news is that you are now a believer in in-memory databases," said Sikka, claiming a four-year head start. "We have already changed the game and we are moving onto the next frontier."

That's brave talk, but Hana deployments are still measured in hundreds of customers, so the market opportunity hasn't changed much. With both Oracle and Microsoft now promising in-memory performance from the most popular databases in the world, SAP is likely to see the wind taken out of Hana's sails (and sales).

Yes, Oracle and Microsoft have yet to actually release their in-memory products and their performance has yet to be tested. But most customers don't seem to be racing to get off their incumbent databases.