Back in the early days of computing, when I worked on RCA, Burroughs, Univac, Honeywell and IBM mainframes, all of them had the same basic model of data in play.
Data belonged to applications.
Through all the changes since the late 1960s-early 1970s, for most of us, this is still how it is.
This data belongs to that package.
Along the way, we added data warehouses to the mix, and then business intelligence tools to parse those holdings. But the elements still went back to applications.
If you want to know why getting to a unified information architecture with a master data model has been so difficult, there’s the answer.
One interesting approach seen of late to deal with this transition is the concept of a “data bus”.
Now this is an idea: underneath it all, we have standard databases holding standard data objects as before.
But the idea is to put in place a flow from one piece of the enterprise’s puzzle to another.
Cycles and storage are cheap these days, by comparison with the past. Some of them are used to transform raw data into intermediate data objects that define the information architecture.
Change the package or custom application that creates the raw data? No problem: just write a new transform routine. The intermediate data object is inviolate.
Another type of routine running on the “bus” (I say “running on the bus” as these are triggered to do their job through message passing, and can operate in parallel to handle loads) is the populating routine. Here, the “standard data object” for a common piece of information — say the definition of a supplier — is populated into intermediate data objects that need a supplier reference.
These, in turn, trigger any application with supplier references to be updated to use the latest common supplier information. It doesn’t take long for the referential integrity of elements to unify across all the applications in an enterprise that use them.
Every application “plugs into” the bus, getting inputs from it, and supplying outputs to it. The bus (the databases of intermediate objects) act as the way to allow change to occur without requiring changes to propagate elsewhere.
It turns out in practice that the amount of processing power and storage space required to put in this isolation and standardization layer isn’t as large as you’d think, especially since there’s no requirement for this layer to be domiciled in the same technology framework as the applications. High-powered Linux solutions — perhaps an old IBM z/System mainframe with its high I/O capability — might be used to do transforms, even if all the applications run on Windows-based servers.
The data bus concept has multiple benefits: data quality improves dramatically, key information becomes univocal (the goal of a master data model), it is extensible, allows for quality control and auditing (transforms make clear what is present in the middle of processing processes), securable (information provision externally to the enterprise, or around the enterprise, draws upon intermediate objects whose content is controllable, preventing leaks of sensitive data) and makes a higher rate of change possible. The use of lowest total cost to own and operate (TCOO) technology, in turn, helps set the stage for the migration of other platforms to lower TCOOs as well.
What the data bus illustrates is that today’s IT world shouldn’t be thought of in the ways yesterday’s was. Part of the transition from IT mostly being about technology and only incidentally about information to the converse — information-centric, with technology serving that — is that we must design data and information, then plug applications into it.
Vendors — of software-as-a-service, of business process outsourcing, of packages and of infrastructure — will find a data bus-centred organization a less congenial place to come and extract money from, as the demands to fit into this world will not be part of their standard thinking. Systems integrators, on the other hand, if they can be made to see what is wanted (as opposed to delivering another clone of a prior project), would find this market more than congenial.
For enterprises seeking value generation from information, the data bus is a good (and, likely, essential) step in the right direction.