Teradata Universe in Warsaw
Is data warehousing writ large an opportunity for developers?
After a couple of days with the Teradata people at its Universe Conference 2007 (22-25 April), I'm seriously impressed.
I like Teradata's "one data warehouse" concept, with all the disaggregated data in a normalised store. I like the way it separates the logical views of the data (including the semantic view) from underlying physical database.
And, I think I even understand the use of hashing rather than indexing to get at the stored data and why it is faster (basically, you hash the key to get a unique location on disk, which is a bit less efficient than just adding it in the next available space in an indexed database, but thereafter you can go straight to the data without any indexing overheads) – because that is how IBM's IMS HDAM databases work (albeit IMS is optimised for transaction processing rather than retrieval).
Teradata chooses to market to the largest companies in any space, so its installations are usually big and expensive, but it can scale down – in Finland, for example, it might have customers with comparatively small data warehouses, because there aren't that many people in Finland.
There is also a flourishing developer community and I hope to explore this more in the future. Could there be advantages for a developer moving into a comparatively specialised space such as Teradata application development?
The general theme of the conference is "embracing active enterprise intelligence" - and an excellent point was made that fact-based decision-making often needs near-real-time information and that the real metric for any data warehouse is the active use of its data to make decisions
An interesting announcement at the conference was "multi temperature" data warehousing – automatic management of "hot" (frequently accessed) data and "cold" (less accessed) data in the same warehouse, with management in logical (SLA) terms instead of physical terms.
So, the data you need for near-real-time operational decisions) might be on expensive "enterprise quality" disks and the stuff kept just for contingency regulatory reports on something more commodity level – and this would be transparent to the data users/owners who just specify the appropriate service level. This doesn't handle migration to archive/backup yet, but it might in the future.
This is interesting and (as far as I can see) new to data warehousing. Although, of course, IMS implemented something similar in the OLTP space years ago and IBM has had logical object-based data migration tools for some time.
Another interesting development is a partnership between Microsoft and Teradata, so that Microsoft's excellent BI tools can be used against Teradata data warehouses. I did ask if this was a tacit admission that SQL Server couldn't scale and was told by Microsoft (I paraphrase) "not exactly".
But Mark Whitehorn points out that Teradata's Inmon-style single warehouse appears to be in conflict with SQLServer's data mart approach (which neatly avoids any scalability issues, for a while at least) and I've asked him to explore this in a future article.
One more thing to remember is that Teradata is splitting off as an independent company from NCR soon, and that might generate new opportunities. On the whole then, an informative conference – and Warsaw's hospitality was excellent. I particularly enjoyed the mezzo recital Teradata provided at its Opera House gala. ®