As an early pioneer of data marts, Informatica saw that large-scale data warehouse projects would be unmanageable for most customers. So the company struck out on its own last year to deliver a client/server data mart offering, called Power Mart, which allows individual departments to create their own data marts. Now the company is about to deliver its Power Capture middleware to link data marts. Informatica CEO Gaurav Dhillon talked with IDG's Michael Vizard about the key issues facing data warehouse implementers.
IDG: Today people seem to be moving toward the concept of data marts, which are more targeted versions of data ware-houses aimed at departmental applications. What's driving that change?
Dhillon: The large initial implementations - we call them galactic warehouse approaches - have not had as much success as have the more bottom-up kind of approaches, which are more characterised by something that is a data mart, where you're looking at essentially a subject-oriented, functional kind of database for decision support. The chief merit of this approach is being able to deploy that quickly.
IDG: But people say that when they deploy data marts, the thing expands and scales up. That puts them right back where they were, trying to manage the big data warehouse.
Dhillon: If you don't architect data marts properly, that can happen. The notion of having an architecture in place where you can roll out multiple data marts is a very important one. What we've always espoused is the enterprise data mart concept, where you have an architecture into which you roll out multiple data marts and provide the ability to have them work together. That means being able to manage them from a single point of control, being able to scale up the data-volume sizes, and being able to handle enterprise kinds of problems.
IDG: What does Informatica provide that makes it easier to manage multiple data marts?
Dhillon: The Informatica product has been designed where one can use, for example, our management capability to run sessions or mappings of data movements from various operational systems into separate databases from one point of control. You can go out and control a data mart from a remote location, or you can manage multiple data marts. This also has to do with not only the building of a data mart, but the care and feeding of data marts, which is equally important; you just build it once, but then it has to be refreshed every night or every other day. Being able to care and feed for multiple data marts using a single point of control is very, very important.
IDG: How hard is it for the average IS shop to build a data mart?
Dhillon: Fairly easy. The way we built the products is that they are designed for the client/server programmer, as opposed to a database administrator. They can go out and deploy this installed from a single CD-ROM.
IDG: What the average manager would want to see is a rule applied in the database that, if this rule is violated, an agent would act by pushing something out to somebody to tell them to take an action. How close are we to that?
Dhillon: The approach we've taken is called exception processing. I think the difficulty of these situations is that a human basically knows how to react, so you don't have this kind of programmed overreaction to something that could be fairly benign.
We've developed a capability to handle exceptions of varying degrees of intensity. So, for example, if you've loaded a certain amount of rows, say 900,000 rows, and the row that's 900,001 fails, should you reload the entire thing or should that be handled in such a way you say 'Look, log that exception and carry on. I will come back and deal with that in time'?
If you have a hard disk fill up, this is probably a disaster where you should beep somebody or at the very least send them e-mail and bring this thing to a graceful halt. We have rule-based processing in the engine and the ability to handle varying degrees of exceptions. So it's not as fancy as avatars and so on, but I think the notion is that systems management will move in that direction.
IDG: How will this be accomplished?
Dhillon: There's a notion of being able to extend meta data capabilities to be broader and across the enterprise. So we'll be extending what we do in our meta data. We will really start extending that by providing a better solution to customers using what we call the 'exchange initiative', where you can provide a single unified point of access for meta data.
IDG: What sets Informatica apart from the rest?
Dhillon: We found that prior technologies had very much of a mainframe feel to them and were very expensive. We started this whole notion of data marts as being a bottom-up way of approaching these projects.
The other notion of what we have as a very central theme in the company is that the bulk of the market is still building databases by hand. Therefore, they have no meta data, they have application development backlog, they have systems management issues when it breaks, and there's no way to handle exceptions in a graceful way.
So the ability to provide a foundation in which you can build these enterprise data marts effectively is a vast and untapped market.
Informatica doesn't have an Australian agent or distributor yet, so contact:
240 Constitution Drive
Menlo Park, CA 94025
Tel (415) 462-8900 Fax (415) 462-8910