Oracle Moves to In-Memory Analytics With Exalytics Appliance

 
 
By Chris Preimesberger  |  Posted 2011-10-03
 
 
 

SAN FRANCISCO -- Oracle OpenWorld 2011, like most mega-conferences of its type, is all about potential customers, analysts, integrators, partners and developers asking three main questions: a) What's next; b) how much; and c) when?

In Oracle's case, the answers this year are: a) Exalytics; b) undisclosed at this time; and c) undisclosed at this time. So how's that for getting the news facts together?

The big picture here is that Oracle now claims to have completed a fast, DRAM (dynamic memory) and NAND flash-based, big-data-type analytics server that runs its own software or that of Apache Hadoop or EMC Greenplum.

Never mind that at this point it's still vaporware to most people and that these things, when they do actually become widely available, will contain the most expensive components in storage. Never mind that once you buy into the Oracle world you're pretty much committed for life. Putting all those hesitancies aside, it's fun after all to think about how fast these might be and how much work they might be able to do in short spans of time.

Some Details Now Available

We do know some details about Exalytics (left). This soon-to-come (most likely by early 2012) in-memory appliance is Oracle's latest whiz-bang box full of IT goodness. Exalytics moves "at the speed of thought," is how CEO and co-founder Larry Ellison described it in his opening keynote Oct. 2 at the Moscone Center.

Exalytics is comprised of a Sun Fire X4470 M2 server, which is a four-socket, 3U-size box running Intel's multicore (in this case, 10 CPUs) Westmere-EX Xeon E7 processor. As such, it features 40 processors. Each box can store an impressive amount of 5TB to 10TB of deduplicated, compressed data in memory.

The secret sauce, however, begins with the solid-state innards: a full 1TB of DDR3 main memory and six flash disk drives packed onto the chassis. It features 40G bps Infiniband connectivity with optional 10GbE ports.

Oracle's parallelized TimesTen relational online transaction processing (OLTP) and Essbase parallel online analytical processing (OLAP) databases are what processes the Big Data workloads.

Travels at the 'Speed of Thought'

"This really is what we consider 'instantaneous in-memory response,'" Ellison said. "You cannot get any faster than this. The Essbase analytics are tuned specifically for in-memory processing. These things work basically at the speed of thought."

You have to hand it to Ellison for coming up with a catchphrase virtually every year at this conference, which has about 45,000 registrants this year. Or at least we should credit his speech writer.

Exalytics can scan an in-memory database at speeds of up to 200GB/sec. "To deliver data analytics at the speed of thought - faster than you can type - you have to change the interface because it moves so fast," Ellison told some 3,000 attendees Sunday night. "Before you finish asking a question, it can guess and give you the answer. Think of much more work you can get done because you can ask so many more questions."

Exalytics also parallelizes (try saying that word fast five times) analytics algorithms so that they run optimally across up to all 40 CPUs as needed, Ellison said.

Naturally, the Exalytics connects natively with the previous Oracle servers and storage arrays based on Sun Microsystems-created IT: the Exadata OLTP/data warehouse appliance (through an Infiniband link), the Exalogic database server (10GbE), all of its Sun and StorageTek disk and tape storage arrays, and all its EMC-affiliated storage.

Oracle OpenWorld 2011 continues through Oct. 6 at the Moscone Center. JavaOne, being held in conjunction, is at the Hilton.

Rocket Fuel