Over the last couple of years, Teradata has extended its family to include a range of data warehouse appliances. Most recently, on Oct. 19, the newest addition, the Teradata Extreme Performance Appliance (code-named Blurr), was announced. This appliance uses all SSD (solid state disk) storage and cranks out a very impressive 55,000 IOPS/sec (input/output operations per second). It won't be a cheap purchase, but you'll get twice the price performance offered by traditional rotating disks.
On first acquaintance, it seems a very odd area for Teradata to embrace. Teradata has adopted a firm position about how business intelligence (BI) systems should be built. In very broad terms, most of the rest of the BI world believes that data should be pulled into one place (a hub) and subsets of the data to be used for specific purposes should be pulled out into data marts, aggregated and made available for analysis. This is the so-called 'hub and spoke' model, the marts forming the spokes.
Teradata is pretty much unique in its approach to delivering the same functionality: It says that all data should be put in one place known as the Enterprise Data Warehouse (EDW) where it should be held in a relational store and all analytical questions run against this central store. This approach brings huge advantages: It makes the move toward real-time analysis much easier, and it enables Teradata to implement its Active Data Warehouse where BI can be used not only to simply analyze transactions that have already taken place but to report on transactions while they are taking place. For example, it can be used to give the probability that a given sale is fraudulent as that sale is taking place, allowing the vendor to decide whether to complete the sale or not. One important characteristic of this model – there are no data marts.
So what is Teradata doing playing around with data marts and data warehouse appliances? At worst, it could be seen as an admission that the company's entire unique model of data warehousing is fundamentally flawed. Actually, the fact that Teradata offers both marts and data warehouse appliances is entirely logical within the context of the real world in which Teradata operates.
There are three reasons why these options are to the advantage of Teradata's customers. One is simply that there are some data sets that need to be analyzed but which should not be in the EDW. For example, the EDW should contain the data that the business as a whole needs to analyze -- customer, sales and employee data. Meanwhile, the IT team wishes to analyze sales by network traffic. In theory, you could put the network traffic data (which will be a huge data set) into the EDW and run the analyses. This would be very wasteful of resources as the network traffic data is only ever likely to be of interest to the IT group. It makes far more sense to use the EDW to provide one version of the truth about sales and to pull this out into a data warehouse appliance, thus preserving the consistency of customer data with the rest of the organization. The huge network traffic data set is also loaded into the DWA and the IT group's analyses can proceed without affecting the EDW's performance for all other users.
The second reason is maturity. Not all of Teradata's customers have reached the stage where a full EDW has been implemented. These organizations may need data marts and data warehouse appliances as they move forward toward that goal.
The final reason is politics. Groups within an organization insist on working within a silo and having complete control of 'their' data. This requirement can be met by implementing a data warehouse appliance.
Teradata offers both marts and appliances because some of its existing enterprise customers have a need for these, and Teradata is delighted to supply the goods. It seems unlikely that Teradata sees itself selling data warehouse appliances to very small companies in competition with the likes of Cognitio and Greenplum. Customers who might have looked toward other suppliers now have the option of staying on known ground and adopting a further Teradata solution.
About the author: Dr. Mark Whitehorn specializes in the areas of data analysis, data modeling, data warehousing and business intelligence (BI). Based in the U.K., he works as a consultant for a number of national and international companies, designing databases and BI systems. In addition to his consultancy practice, he is a well-recognized commentator on the computer world, publishing articles, white papers and books. He has written nine books on database and BI technology. The first one, "Inside Relational Databases" (1997), is now in its third edition and has been translated into three languages. The most recent is about MDX (a language for manipulating multi-dimensional data structures) and was co-written with the original architect of the language, Mosha Pasumansky. Mark has also worked as an associate with QA-IQ since 2000. He developed the company's database analysis and design course as well as its data warehousing course.