Exadata v2 Raises Questions about Data Warehouse Market
IBM's rumored announcement of an "Exadata Killer" turned attention to Oracle's latest powerhouse. Why is there a surprising paucity of on-the-record information about what enterprises are using Exadata and how they're using it?
- By Stephen Swoyer
- October 14, 2009
At press time, rumors were swirling about a forthcoming announcement of an "Exadata Killer" from IBM. The question is -- what kind of a market is there for these uber-Big Four data warehouses? That question can perhaps best be answered by taking a look at the first uber-database machine: the Oracle Database Machine, now called Exadata, which Oracle announced last September.
Central to the discussion is knowing how many Exadata (nee Database Machine) customers Oracle Corp. actually has. No one seems to know, and Oracle isn't saying. It's is dropping broad hints, but when it comes to specific data about Exadata popularity and use, there's a surprising paucity of on-the-record information.
Dropping those hints, of course, is Oracle chief Larry Ellison, along with other company officials. In Oracle's March earnings call with financial analysts, for example, president Charles Phillips claimed that "we are getting great feedback from our customers and the [Exadata] pipeline is the largest build I've ever seen in terms of a new product."
Phillips invoked the example of a "major European retailer" that had used Exadata to achieve a 16-fold reduction in batch-processing time. Phillips' boss was even more enthusiastic. During the same conference call, Ellison called Exadata "the most exciting product [Oracle has] had in many, many years."
He reprised this theme during a June call, describing Exadata as the "most exciting and successful new product introduction in Oracle's 30-year history." This time, Ellison gave financial analysts a bit more to chew on, citing customer wins "in a couple of different smart-phone manufacturers" (one of which he identified as Research in Motion, or RIM), along with Amtrak, Thomson Reuters, and "the largest Teradata customer in Japan." In addition, he mentioned up Barclays Capital, unspecified customers in Korea and other Asia-Pacific locales, and "a number of banks in West Germany" as satisfied Exadata customers.
Comparatively few customers seem willing to talk about their use of Exadata. An Oracle Exadata v1 promotional video points to a trio of reference customers -- viz., LGR Telecommunications (which markets reporting and analysis services to telco customers), the Chicago Mercantile Exchange, and M-Tel (a telecommunications provider that actually did discuss its Exadata experience with Intelligent Enterprise magazine) -- but prominent, named customers, such as RIM, Amtrak, or Thomson Reuters, were MIA. Nor did Oracle showcase any reference customers at its recent Exadata v2 launch event.
Just how many Exadata customers are there? Analytic database watcher Curt Monash recently posted the blog equivalent of an Amber Alert on his DBMS2 blog, posing just such questions.
"[C]an anybody identify much in the way of Exadata production references? Oracle recently talked of a few flagship data warehouse customers, but those don't seem to be running Exadata," wrote Monash. "I talked recently with an Oracle prospect from the U.S., who only got one reference from Oracle -- in Eastern Europe." The upshot, Monash went on to suggest, is that "Oracle Exadata production sites are pretty scarce on the ground."
That's a take echoed by others in the analytic database segment. Consider columnar database specialist ParAccel Inc., which likes to pitch its ParAccel Analytic Database (PADB) as an accelerator for performance-plagued Oracle environments.
According to CEO David Ehrlich, ParAccel almost never sees Exadata in the field. "Larry's a phenomenal marketer," says Ehrlich, citing Ellison's claim that Exadata is both faster and cheaper than "specialty" analytic databases such as Teradata, Netezza, or (although didn't Ellison mention it) ParAccel, "but [Exadata is] something we just don't hear about [from customers]. I don't think we've ever run into it in a competitive situation."
The same is true for Aster Data Systems Inc., an analytic database specialist that markets a DBMS-native implementation of the popular MapReduce API.
In a sit-down interview at this summer's TDWI World Conference in San Diego, Raj Pai, vice president of strategic alliances with Aster, stopped short of dismissing Oracle and Exadata as a competitor. "I don't really know offhand whether we've encountered it [Exadata] in a sales situation or not," Pai conceded. On the other hand, he said, Aster does see a lot of the vanilla Oracle database; many of its customers are disaffected Oracle shops which deploy Aster's nCluster database to address chronic Oracle performance problems.
"What we've found primarily is that in our pipeline prospects as well as the customers that have now put [nCluster] into production, the key pain where they initially engage with Aster is really a … lack of scalability for their data needs. Whether it's starting out with MySQL, SQL Server, or Oracle, they're seeing a proliferation of data -- primarily [of] electronic or digitally-driven data," said Pai.
"I think they're happy with their choices of these databases for OLTP, … but what they've found is that they've hit a bottleneck when it comes to doing queries and analytics. This is as true for Oracle as it is for MySQL or SQL Server."
Wouldn't shops that have significant investments in Oracle -- as both an OLTP and as a decision-support platform -- be especially receptive to its pitch with Exadata? If they aren't, Pai suggests, it's probably because comparatively few shops -- particularly in the kinds of large organizations that tend to be candidates for data warehouses with large data volumes -- are all-anything environments.
It isn't just that they need a way to accelerate Oracle analytic performance, he argues; it's that they frequently have other DBMS platforms in-house, too. It's a tendentious issue, of course, because just as some customers are juggling a mix of workloads running on at least two platforms (such as an Oracle customer in the UK –-- see http://www.tdwi.org/News/display.aspx?ID=9143), others have built large data warehousing practices on top of off-the-shelf DBMSes -- usually Oracle -- and are now running out of headroom. Hence the interest in Exadata -- or (Pai suggests) search for analytic databases such as nCluster.
There's also the question of Exadata's price tag, with which seemingly all of Oracle's analytic competitors take issue. By all accounts, Exadata v1 was a premium offering: Oracle initially fixed its cost at around $14,000 per TB, but industry-watcher Monash famously reckoned that Exadata -- outfitted with Oracle's all-but-essential Real Application Clusters (RAC) option -- actually clocked in at either $60,000 or $130,000 per TB, depending on if one opted for SATA or SAS storage.
The lesson, of course, was that Exadata v1's too-good-to-be-true pricing -- which was, relative to the $100,000-per-TB advertised prices of vendors such as ParAccel and Vertica Inc. (and even in comparison to the $60,000 per TB advertised by Netezza), obscenely affordable -- was, in fact, too good to be true.
Last month, Ellison claimed that Exadata v2 was both faster and cheaper than any other OLTP or analytic competitor on the market, but Oracle has had to backtrack from at least one of Ellison's claims -- namely, that Exadata v2 offers better price/performance than the reigning OLTP champ, a System p-powered RISC-Unix system from IBM Corp. Ellison's contention, it seems, relies on an as-yet-unaudited TPC-C benchmark, which drew the attention of the Transaction Processing Performance Council (TPC). The organization levied a $10,000 fine against Oracle and directed that it remove an ad (which ran in The Wall Street Journal and other venues) touting the performance of Oracle's 11g database running on top of hardware from Sun Microsystems Inc.
To be fair, Oracle isn't the only vendor to have pulled a TPC about-face. ParAccel, for example, recently withdrew its contentious TPC-H benchmark result following a challenge by a competitor and a review by the TPC. "We published our results in good faith," wrote ParAccel's Dave Steinhoff on his blog. "However, the body of TPC rules is complex and undergoes constant interpretation; we are still relatively new to the benchmark game and are still learning, and we made some mistakes."
Exadata in the Wild
Tim Young, vice president of corporate marketing with Netezza, has encountered Exadata in the field. More to the point, Young concedes, Exadata sightings aren't all that rare -- and Oracle has even used Exadata v1 to bloody Netezza's nose, albeit, Young stresses, in a strictly immaterial way.
"We have encountered [Exadata]. We've encountered it in about 10 situations, [and in] the vast majority of those situations Exadata has not succeeded over Netezza," Young says. "There is at least one situation I know of that Exadata was selected over Netezza. The selection was not based on whether the product was the fastest or the best or the cheapest; [instead] it came down to a clever bundling agreement that Oracle was able to do."
It's this ability to exert pressure (or offer carrots, depending on your perspective) in a variety of contexts that makes Oracle a formidable foe, Young argues. "This is what makes Oracle a difficult competitor. Oracle's technology doesn't have to be great. Oracle can become a significant competitor if their technology is being perceived as good enough, and if people buy Oracle as part of a bigger reason," he maintains. "Oracle has a lot of variables that they can play with," Young continues, citing Oracle's formidable software and -- assuming the Sun deal goes through -- hardware portfolios.
How many Exadata customers are there? At press time, Oracle had not responded to repeated requests by BI This Week for comment and clarification. Monash has done some digging, and Young thinks he has a good ballpark idea, too.
On his DBMS2 blog, for example, Monash relays the anecdotal account of a "classified-agency user" that's hosting more than 1 PB of data on Exadata, along with an additional 600 TB on Netezza. He's more critical of a rumor that Infiniband switch purveyor Voltaire Ltd. has recorded "triple-digit" sales of switches into Exadata environments (presumably, one per installation), citing a lack of corroborating evidence (e.g., in earnings reports, conference call transcripts, or SEC filings). "[T]he most recent transcript does seem to indicate Voltaire got multiple Exadata deals in the telecommunications sector, and suggests some Exadata penetration in other sectors as well," he concedes.
Young, for one, thinks the actual tally falls considerably short of triple-digit territory. "My own sense, with all of the analysis that we've done and all of the conversations that we've had with customers … our guess is that Oracle [probably has] 25 or so installations for Exadata, and I suspect that a good number of those are not paid-for installations," he comments. "There are [Exadata] partners who they quote in Japan who are also our partners, and Oracle is quoting these as being exclusive [Exadata] customers when in actual fact these are demo environments."
If anything, Young claims, Exadata has been a net positive for Netezza. "I can say that in the same period of time since Oracle announced Exadata, we've had over 50 Oracle customers churn for Netezza. Those customers would all have been given the briefing by Oracle, they all knew that Exadata was on the horizon, they all would have sat down and tried to make it [Exadata] work. Because, quite honestly, if you can stick with your current supplier -- that's the most natural thing to do," he continues, citing Ellison's concession in a (June 2009) Oracle earnings call that Netezza was "kind of tied" with IBM (behind Teradata) as Oracle's number-two competitor. "Those 50 organizations decided that even though Oracle is their preferred supplier for the database, they're going to go with the new kid on the block in Netezza."