Teradata’s Active Enterprise Data Warehouse story
Teradata used to tell a one-size-fits-all Enterprise Data Warehouse (EDW) story. That’s no longer the case. Last year, Teradata introduced a range of products. I think Teradata is serious about selling its full product range, and by now has achieved buy-in from its sales force for that strategy. I base these beliefs on data points such as:
- Teradata says so, repeatedly and persuasively.
- At least in passing, Teradata cites non-trivial sales figures for the appliance product lines.
- Competitors are less unanimous in asserting that Teradata’s lower-end products are presented on just a bait-and-switch basis.
But that raises the question: How does Teradata pitch the advantages of its top-end product line these days? At least at the corporate level, the answer seems to focus less on the “EDW” concept than it used to, and more on “Active.” Teradata – which actually has been talking about “Active Data Warehousing” for about a decade — indeed calls its top-end 55xx series the “Teradata Active Enterprise Data Warehouse.”
Teradata proudly told me that it has >100 customers who have truly adopted an “Active” EDW. When we discussed what that meant, supported by a whole lot of named examples, it became clear that “Active” data warehousing:
- Has a lot in common with “operational” analytics
- Is most often but not always used in customer-facing applications
- Has three main technical themes:
- Low latency
- High concurrency
- Robust uptime
Business rules and/or “exception-based” processing also seem to play a significant role, which is of course a good fit for the “operational” theme.
Other kinds of use cases cited included:
- Transportation and logistics
- General BI at Teradata itself and former parent company NCR
Also, Teradata was able to quickly cite examples in both social network analysis and anti-fraud — but not in the use of social network analysis for law enforcement or fraud fighting.
How low is the latency? Well, the lowest-latency use case I recall is Travelocity’s, which believes the queries supporting web personalization need to resolve in 300 milliseconds or less, with >2 second response being a “Failure”. But figures as low as 15-30 milliseconds got bandied around in conversation. At the other extreme, I’d guess plenty of customers in that >100 count have latency requirements no more demanding than a few minutes; in some cases perhaps even a couple of hours would be OK.
Anyhow, we’re not talking about CEP (Complex Event/stream Processing). But at least for the customer-facing apps, we are indeed talking about fast query response. And , in the one part of the pitch that really does sound like the old EDW story, Teradata stresses that this fast query response should occur even in the face of mixed workloads — i.e., Teradata makes a case for the relevance of strong workload management. Not coincidentally, workload management is the “only” software difference between Teradata’s “Active Enterprise Data Warehouse” 55xx series and the rest of the Teradata product line.
Of course, Teradata isn’t alone in telling the “Active” data warehousing story. Active data warehousing – although not necessarily by that name – is where HP tends to focus Neoview. It also is a lot like what Aster Data calls “Frontline” data warehousing, although Aster generally puts more emphasis on uptime and less on concurrency. And anybody telling a one-size-fits-all DBMS story – e.g., about Oracle, Microsoft SQL Server, or DB2 – will surely cite “Active” concepts at times as well.
Comments
6 Responses to “Teradata’s Active Enterprise Data Warehouse story”
Leave a Reply
[…] couple of slides in my recent briefing onTeradata’s Active Enterprise Data Warehouse Story contained long lists of analytic use cases, at a finer level of granualarity than I’m focusing […]
A small clarification — there are two latencies being discussed in this blog.
First is the response time for web site (eCommerce) fast queries which typically run 1-5 seconds. Residency time for a unique key read is less than 40 milleseconds on the Teradata 5500 series. However total round trip from client, TCP/IP internet, to web app server, to the RDBMS and back has many possible speed bumps. So while 1 second queries are common, some may creep up to 2-4 seconds.
The second latency discussed in the blog described as minutes and hours are for mini-batch loading data concurrent with complex queries and operational queries. So loading a few hundred megabytes every 15 minutes 24X7 is becoming common with Teradata servers. Latencies < 2 minutes are in production but only a few applications can justify the labor and complexity involved so 15 minute mini-batch is more common.
Thanks Curt — a good unbiased write up.
Dan,
Thanks for the additional info!
Curt- Good initial struggle with defining ‘active’ data warehousing. Instead of getting involved with confusing factors, like latency, think of the concept of ‘active’ as an effective Stimulus->Response->Stimulus cycle within a business. First, you need to know what is happening. Second, you need need to respond (capture, store, analyze, decide, act) to the situation. Third, you need to make an impact. Do that quickly and smartly, and you are ‘active’. -Richard
Richard,
Then “Active” just becomes a performance review for the analytics team. If they are Smart and build Quick systems, then their warehouse is active. If they are Slow and build Dumb systems, however, then it isn’t.
[…] recently wrote that Teradata had gotten serious about appliance product lines, and had non-trivial sales figures for them. In a press release today, Teradata is now explicitly saying (emphasis […]