ODS: For Flexibility and Speed
by Bill Inmon
Originally published July 8, 2004
If ever there was stealth technology, it has to be the ODS (Operational Data Store). ODS has been around for years. Corporations usually have many of them. But, strangely enough, there is no support for ODS in the public sector. There are no conferences. There are no seminars. There are no classes. There is only one book ever written on the subject - Building the Operational Data Store.
In short, if you want help with your ODS, you will need to find it or create it on your own.
There is some vendor support coming from SAP and PeopleSoft. Both SAP and PeopleSoft recognize the ODS as a prominent feature of their BW and EPM architectures. But other than that, there simply is little or no support from the vendor community for ODS.
This lack of public support is remarkable compared to other architectural structures. Data warehousing and OLTP each has its conferences, books, seminars and vendor products. E-Business has its industry support. Portals have their industry support. CRM has its industry support, and so on. It seems that each of these applications has its own supporters but ODS has none. And the ODS is pervasive. It is everywhere the data warehouse is. It is the true Rodney Dangerfield of technology – it just gets no respect.
An interesting question is – why has ODS failed to gain industry support, despite being so pervasive? There are probably many reasons for the lack of support, but the best reason is the complexity of the technology required for an ODS. The nature of the ODS is the blending of both OLTP response time and integrated decision support processing. Anyone who has faced these problems knows that they are akin to walking on a high wire over Niagara Falls trying to balance twelve pizzas and make the pizza delivery in time to get a tip. It is a real balancing act requiring all sorts of skills, compromises and tradeoffs.
Because there is a scarcity of vendor products in the ODS marketplace, unless you want an SAP or PeopleSoft ODS, your only available choice is to build one. And that is exactly what many corporations have done. But the lack of vendor support is changing dramatically. Finally, there is a company and a product in the marketplace that “gets it”. They are CenterBoard in Pleasanton, CA.
CenterBoard is discovering that there is a marketplace for a virtual ODS. Some of the features of CenterBoard that make it attractive for the organization that wants to build an ODS are:
CenterBoard produces query results that can be called “temporal” results. A temporal result is one that is accurate only at that moment in time. I make a query at 2:56 pm and find that corporate revenue is $X. At 3:01 pm I make another query and now revenue is $Y. In an ODS that discrepancy of values is absolutely normal. The reason is that the ODS operates on changing values of data. The accuracy of a query in an ODS is only at the moment in time that the query is made. Any other assumptions made by the user are simply incorrect.
Because of the temporal nature of an ODS, creating a “virtual” ODS, or even a “virtual” query is an acceptable thing to do. Unless you are going to be doing repeated queries against an ODS, you do not need to actually create a physical ODS. One major benefit of creating virtual ODS queries is that there is no need for moving the data. Data can be accessed “as is”, in place. (Or as golfers like to say “play it where it lies.) The ability to manipulate data in place speeds the time in which queries can be done and greatly lowers the cost of queries.
One of the great advantages of a “virtual” ODS is that it is so flexible. One query can look at one thing and another query can look at something else. There is no infrastructure that has to be rearranged every time the requirements change.
In addition, CenterBoard supports queries that are the architectural equivalent of an oper mart. An oper mart is to the ODS what a data mart is to the data warehouse. The oper mart is used for analytical processing. But there is an essential difference between an oper mart and a data mart. In a data mart there is stability of analytical values. If you need to rerun a query or an analysis against the data mart in the exact form in which it was previously run, the query against the data mart should be repeatable. (If you can’t do that then you have built your data mart incorrectly.)
But with an oper mart, you can still do analytical processing as you did in the data mart environment. However, in an oper mart you cannot expect repeatability of results. Data is subject to change at a moment’s notice in the oper mart. So the second time you run a query against an oper mart you may be running on different data. Other than that, data marts and oper marts are pretty much the same.
So CenterBoard not only supports ODS, but supports oper marts as well.
It is about time that the ubiquitous ODS got some support. Now organizations have a place to turn to when it comes time to build the next ODS. (Which surely beats the alternative of having to build it yourself?)
SOURCE: ODS: For Flexibility and Speed
Recent articles by Bill Inmon
Copyright 2004 — 2020. Powell Media, LLC. All rights reserved.
BeyeNETWORK™ is a trademark of Powell Media, LLC