As the race to in-memory computing enters the first turn, some sharp elbows are starting to be thrown.
At the recent Oracle OpenWorld 2013 conference, Oracle outlined an in-memory option for the Oracle 12c database.
According to Oracle CEO Larry Ellison, the design goal for this option is to make queries run 100 times faster within both analytical and transaction processing applications by simply switching a switch in a way that requires no changes to the SQL application. In fact, Ellison says the in-memory data option will also process transactions twice as fast using a hybrid in-memory database architecture that stores data in both row and columnar formats.
In addition, Ellison notes that customers are not required to put all their data in memory or disk. They can opt to deploy a subset of the database in memory or in disk, and if they opt for a scale out architecture they can deploy the Oracle Exadata database appliances.
That announcement immediately drew responses from rivals that chided Oracle for everything from not having attached an actual release date to the availability of the in-memory option to pointing out the fact that Oracle now has no less than three in-memory computing platforms in the form of the new Oracle, 12c database, the company's existing Exadata appliances, and a TimesTen database that the company acquired way back in 2005.
Taking that criticism a step further, Sean Poulley, vice president of databases and data warehousing for IBM, contends that the IBM BLU Acceleration platform is a more unified approach to combining in-memory and magnetic storage that more seamlessly integrates data across different storage paradigms. Similarly, Teradata has taken an approach that looks to seamlessly integrate in-memory computing and magnetic storage in the form of Teradata Intelligent Memory.
In contrast, Poulley says Oracle is providing an in-memory option, while SAP HANA only supports in-memory computing with no option for magnetic storage.
But for all the sudden posturing around in-memory computing by database heavyweights, the hardest punches may soon be delivered by much smaller vendors such as MemSQL, Clustrix, and Actian, which are all castigating IBM, SAP, and Oracle to varying degrees for delivering what amounts to mainframe style approaches to in-memory computing based on legacy architectures that are overly expensive.
"These companies are launching dinosaurs just as a comet is about to hit," MemSQL CEO Eric Frenkiel told Datacenter Acceleration. "The price of memory keeps falling and yet these vendors still want hundreds of thousands of dollars for terabytes of memory in a scale-up architecture."
In contrast, MemSQL is based on a truly distributed scale out architecture that supports standard SQL applications, Frenkiel said.
Similarly, not all approaches to in-memory computing are created equal, John Santaferraro, vice president of marketing for the ParAccel massively parallel database told us. Santaferraro contends that it's not just how well the database makes use of Flash memory, but also the DRAM capabilities of the latest generation on Intel Xeon class processors.
"There's a little bit of doubletalk going on," said Santaferraro. "Big vendors are talking about the price of Flash coming down, but then rolling out systems that cost hundreds of thousands of dollars."
Clustrix CEO Robin Purohit added that IT organizations are tired of having to perform unnatural acts such as sharding to scale database applications.
"IT organizations are voting for horizontal, scale out architectures that one logical database can span," said Purohit. A single server can handle small amounts of data; you need a new approach to handle several terabytes of data running in memory."
While there's no shortage of in-memory database options, Phil Friedman, CEO of the IT services firm CGS Inc., said the best news of all may be that enterprise computing is getting more economically efficient with the passing of each day.
"We're now exceeding Moore's Law many times over," says Friedman. "Cost reductions relative to the amount of processing power required is an unstoppable trend."