Big Box Teradata Takes an Outside-the-Box Approach to Business Intelligence and Data Warehouse
The Teradata of today has an unconventional take on database appliances, in-database analytics, DW virtualization, and data quality, company officials claim.
- By Stephen Swoyer
- May 5, 2010
Two years ago, Teradata Corp. broke with its own tradition and unveiled a line of analytic database appliances.
It was a surprising move by Teradata, which (prior to its appliance about-face) had stuck resolutely to its enterprise data warehouse (EDW) guns. Then as now, Teradata officials like to stress that simply having an analytic appliance strategy isn't in itself inimical to doing EDW.
"I've heard from some customers who say, 'You're confusing me: you guys are the Big EDW [vendor], right? But now you have this platform family. Is the answer now to break everything up?'" says Randy Lea, vice president of product and services marketing with Teradata.
"We'd like you to put everything on the same system, but now we're giving you options. We can sell you an Extreme Data Appliance, an Extreme Performance Appliance, or you can have them both on one [system]. We believe in the long run that the price will be cheaper on one system than on two systems, but we want to give you the choice."
More to the point, Lea claims, some analytic database players (such as Netezza Inc.) are now trying to swim upstream into the uncharted waters of the EDW market.
They're doing as much, he suggests, chiefly in response to feedback from customers. "I think it really emphasizes their [i.e., analytic database vendors] tactical departmental [deployment model] versus more of an integrated or enterprise thing, which is what we have. I'd argue that even if you have one subject area, one department, it isn't as straightforward as that," he explains.
"If you have 14 applications [on a single analytic database], do they all get equal value? These are the kinds of questions [customers that have adopted analytic databases are] dealing with. At this point, [analytic database vendors] don't have any real answers for [customers]."
Quite aside from what he spins as a favorable change in DW deployment, Lea also points to a change in the role of the DW.
"A lot of people have a data warehouse and maybe [between] 10 [and] 40 percent of their data warehouse [workload consists of] doing extracts to cubes, extracts to SAS, extracts all over the place. We want that [to move to Teradata database] and believe that you should be doing that on the EDW," he says.
Teradata has been out in front on this trend, Lea claims, citing his company's strategic partnership with SAS Institute Inc. -- an effort to embed SAS analytics into Teradata warehouse such that the former run in the in-database context of the latter -- as well as other, more recent efforts, such as Teradata's embrace of the open source Hadoop MapReduce implementation.
These moves are of a piece with Teradata's push to change how customers conceive of -- and use -- their DW environments, says Lea.
"Twenty to 30 percent of resources in a warehouse are [used] just [to] move data. That's just wasted resources. The data warehouse is good because it's great data, it's clean data, it's up-to-date data, etc., but many times because of IT restrictions, you can't [do analysis] in the warehouse [itself]. So what happens? You get users saying: 'I'm going to pull [this information] out to play with it myself,'" he explains. "What we want to do is get all of these environments to go to Teradata and eliminate the extracts that are being done."
A Customer-Choice Orientation
In addition to appliances and in-database analytics, Teradata has championed customer choice in other contexts, Lea maintains. He cites new support for running Teradata in either VMWare or in the cloud.
Later on this year, Lea says, his company plans to officially support Teradata-on-VMWare in SMP configurations. "We are going to be releasing VMWare on an SMP box for production with support later this year in the second half," he confirms. "The feedback we're getting there from customers is that in almost any environment, the data warehouse provides a majority of the analytics and information, however … there's Shadow IT Groups out there building their own little data areas.
"Many times, they'll do that on SQL Server, or they'll do that on Oracle or something else," Lea continues. "What our customers are looking at is … I can have several instances of Teradata on that [big SMP] box, and I can provide my users with access to that. This really gives [customers] more flexibility. When they build a new application or new data, they always come to us later on to maintain it, but now it's built on Teradata [on VMWare], and we can bring it into the warehouse a lot easier later -- we can almost bring it in as is."
Lea says reaction has generally been stronger on VMWare than on Teradata's Amazon EC2-based cloud service.
"[Uptake has] been a lot more tactical at this point in time. The volume has been much higher on the VMWare because I think most [adopters] have a VMWare environment already," he says. "One of the areas on the Amazon [service] … we had one customer [that] had a large data set that they wanted to keep and they were keeping that on Amazon, then periodically -- it might even be once a month or twice a quarter, for three days -- they wanted to go analyze that data. We thought this might be a good option for them."
Lea also has something to say about data quality (DQ), which isn't typically viewed as one of Teradata's core competencies. Nonetheless, Lea maintains, customers who focus on data quality at all costs -- especially to the point of constraining DW comprehensiveness or DW performance -- need to consider rethinking their approaches. A DW isn't a Platonic Ideal of clean and inviolate data, he argues. "Data quality definitely isn't [a] fixed [problem]. It's still hard work. I do believe our average customer … has probably as good data quality as anybody. The reason is we do have a lot of resources that … help customers establish good data management processes," he says.
"To a certain degree we have pounded [data quality] into their heads so much that some [customers] go overboard, and they don't let anything in that isn't perfect, and that's a mistake, too," Lea notes, "and we're actually being tagged by some of our competitors, saying Teradata tells you to build this big enterprise model and get it all perfect and … then a year later you get answers. That just isn't true."
Teradata's actual position is considerably more nuanced, he says. "Until you build a warehouse and expose how bad it is, you will never fix your data quality problems. It is this fine balance between cleaning it as best you can [and] once you get it in, identifying what isn't good, go back to the operational sources and people, and … figure out a process to capture [what's missing or what isn't good] and get it back later," Lea concludes, "but you can't identify or fix these problems if you aren't bringing this [data] into the warehouse."