Is There Demand for Exadata?

The trouble with Oracle's Exadata v2 pitch, a competitor alleges, is that there just isn't much demand for what Oracle's selling.

Oracle recently unveiled Exadata Version 2, its most ambitious database appliance to date. Oracle positions Exadata v2 as a jack-of-all-trades database, touting its strong feature set for enterprises with heavy transaction-oriented or data warehousing workloads. (See http://www.esj.com/Articles/2009/09/23/Exadata-Boosts-Speed.aspx.)

The trouble with this approach, one of Oracle's competitors alleges, is that there doesn't seem to be much demand for such an offering.

It's important to consider Oracle's announcement in light of Exadata's short history. Version 1, which Oracle shipped in September 2008, was designed chiefly for data warehousing, delivering what Oracle officials trumpeted as "world-class" performance for analytic workloads.

Oracle tried a different approach with Exadata v2, trumpeting that product's combined analytic and online transaction processing (OLTP) power. Co-founder and CEO Larry Ellison took aim at several competitors in this regard, challenging IBM Corp. -- holder of the existing TPC transaction processing performance crown -- on the OLTP front, even as he dismissed the efforts of upstart competitors (such as Exasol, IBM's own solidDB, and QlikTech) that market in-memory database technologies for both data warehousing and OLTP workloads.

Exadata, Ellison claimed, beats all comers, as both an OLTP and data warehousing platform. He compared Exadata v2 to IBM's Power5-based System p 595, running Big Blue's AIX v5L operating system and DB2 v9.5 database. That configuration, of course, has been sitting atop the Transaction Processing Performance Council's (TPC) TPC-C benchmark heap since last June.

"To deliver the same capacity, the same performance, we need two of these Exadata [v2] racks, which cost less than a quarter as much," Ellison trumpeted. "If you want to spend the same money [as the IBM configuration], you can go four times faster. If you want to save more than 75 percent, just go the same speed," he continued, sneaking in a quick dig at IBM's non-clustered -- and hence non-fault-tolerant -- configuration: "Oh, by the way, we're fault-tolerant; they're not. We're incrementally expandable on demand; they're not."

Nor did Ellison spare Oracle's competitors in the in-memory database segment, either. "We're even faster than this new generation of database that's designed specifically to optimize queries for databases that are stored entirely in main memory," he said. "There's some new start-ups … that all they can do is [have to be] able to load the entire database into memory; if you load the entire database into memory, they run pretty fast."

Not the Real World

Putting aside Oracle's Exadata performance claims -- which, naturally, its competitors dispute, for one reason or another -- the company's rivals seem most incensed about its combined OLTP/data warehousing pitch. Simply put, some allege, Oracle and Ellison are setting up a straw man: the only customers that might actually be persuaded by such a pitch are all-Oracle shops. For most other enterprises, they contend, the costs of running OLTP and analytic workloads on the same platform far outweigh the benefits.

"If you look at most enterprise data architectures, the OLTP applications and the OLTP database sit together and you typically don't run intense analytics on that database. Almost every enterprise has an architecture where they [extract, transform, and then load] the data out of the transaction database into an analytic platform to run analytics on," says David Ehrlich, CEO of Oracle competitor and data warehousing software specialist ParAccel Inc. "The concept that there's a big segment of the market that wants to run their [OLTP] applications and analytics on the same database is just a fallacy."

Ehrlich concedes that such a topology might seem to make sense, but argues that -- in practice -- the costs outweigh the benefits, particularly when it comes to running analytic workloads on what is, pedigree-wise, an OLTP-oriented database.

"Most markets look like a normal curve, and one might think analytics is at one end of the curve and OLTP is at the other, so you might think that the bulk of the people want to do both [on the same machine]. However, because of the way enterprises design their architectures, what you often see is an hourglass distribution," he continues, "so you have a very big market on the transaction-processing side and a big market on the analytic side, but the people doing both of those [on the same platform] is a very, very tiny segment."

ParAccel -- like its competitors -- is by no means a cheap proposition, clocking in at around $50,000 per TB of user data. At the same time, Ehrlich argues, the economics of what might be called the "after-market analytic accelerator" trump those of the combined OLTP and analytic platform. The upshot, he contends, is that customers value Oracle 11g, DB2, or Sybase Adaptive Server Enterprise (ASE) chiefly for their OLTP chops. With this in mind, Ehrlich suggests, shops could potentially "compromise" their OLTP performance by standardizing on a single platform for both OLTP and data warehousing.

Besides, he continues, even if you aren't persuaded by that argument, there's the economic argument: the fiduciary cost of doing as much will in most cases prove to be prohibitive.

"If your bread and butter -- that is, how you make your money -- is on your transactional applications, the last thing you'd want to do is slow down your cash register by running analytics on the same database. Even if [that isn't the case], why would you want to do that when, for a comparatively inexpensive amount of money, you can just offload it into another database that's specialized for these kinds of workloads?"

Industry veteran Mark Madsen, a principal with business intelligence (BI) and data warehousing consultancy Third Nature, likewise isn't persuaded by Oracle's combined DW/OLTP pitch.

"You use Exadata for DW workloads, otherwise why would you buy it? [It] seems to be selling well to the Oracle client base, and it does speed up Oracle BI workloads, so it's a reasonable, if costly, solution relative to the appliance databases," he observes.

"I suspect they're saying this so people can move to Exadata from a regular Oracle install -- for example, not [having to] have two Oracle systems, or maybe to say that you can run BI and OLTP workloads on the same database. That's something Teradata and [HP, with] Neoview were both claiming, e.g. the 'active DW' model."