Data modelling layers: do you wanna get logical or physical
An introduction to data models
Reg Developer: Can't UML be considered a conceptual model? Why do you see a need for a separate conceptual data model?
Donna: Many people do use UML to create abstract, conceptual layer. But while UML does well expressing the interactions between the data and the processes, it does not accommodate some of the core principles around designing data itself and the relationships between data items. Concepts such as identifiers/primary keys, for example, are not expressed in a UML class diagram. A conceptual data model is designed to be easily mapped to a logical data model. UML is more suited to application design [fair enough, but that might just be an omission in MDA, the OMG's Model Driven Architecture - Ed].
We also hear from many of our business users and data modellers that the UML is difficult to understand. Perhaps "difficult" means only that it's not a graphical notation that they are familiar with, i.e. "it doesn't look like a data model to me", or "where's my Visio"? And/or this could be a result of the mistake that many architects make in showing the wrong level of detail to the customer, and that the UML diagrams presented were too low-level to be helpful to business users, and too application-focused to be helpful to the data modellers.
Honestly, another big factor is a societal one. There seems to be separate camps in the UML and ER modelling world and never the two shall speak. This is unfortunate, since I think there are commonalities between the two. I think some of this is changing with efforts like the IMM (Information Management Metamodel) from the OMG, which is attempting to better mesh the UML and ER worlds in a formal metamodel and exchange format.
But, until these two worlds play more nicely together, we'll choose to focus our model on our core audience, the data modeller. But even with that consideration, I do think that the conceptual model as we see it is an easier format for the average business user to understand.
Reg Developer: And XML? How does that fit in? Isn't that data too? I'm sure I remember using data analysis generally and E-R diagramming specifically to design hierarchical IMS databases; and XML is 'just' another data hierarchy.
Donna: XML most certainly fits into these modelling layers. However, similar to the UML discussion, there are currently different camps/factions/silos that don't communicate well. We're seeing our traditional RDBMS data modeller becoming more familiar with XML, but that's a more recent phenomenon. Getting XML developers used to data modelling is a separate challenge - as we mentioned before, the developer crowd can be a harder one to convince of the benefits of a model.
But the benefits are the same - a common conceptual and logical model of the business with different implementation layers - XML being one of those.
We're currently guilty of this disconnect ourselves in our tools. We currently have wizards to export XML from a relational logical design, but we're still treating XML as a translation from the relational world. This actually works well, given the fact that I mentioned earlier that most of our relational database modellers aren't necessarily familiar with XML - allowing them to translate into XML from a format they're familiar with works. But in the longer term, we're working on modelling XML better in its own, native, hierarchical format.
Reg Developer: That's all clear enough, I suppose, and the future directions are interesting, but why do we need a data model at all? Isn't this all just 'stuff' getting between the programmer and her users?
Donna: That's a religious question! There's the basic principal of design, then build. Would you want to live in a house that had been designed without a blueprint?
And then there's reuse - most companies have common data objects such as customer, product, etc. Rather than reinvent the wheel for each new project, better to start from the same core design in the data model.
And what about data quality? In the example above, if every project uses a different definition of 'customer', for example, there are bound to be discrepancies in the data. In fact, there's a whole Master Data Management industry devoted to fixing the problems experienced by companies without a coherent model of their data.
Data governance is yet another issue. The data models help to provide an effective inventory of your data assets. Most companies have a wide variety of database platforms with hundreds, thousands, and even millions of tables. Reverse engineering from these platforms into a physical (and eventually a logical and conceptual model helps companies better understand what they have today – which can be extremely important if these assets form the foundation of regulatory reports, that directors sign off on, on pain of going to jail...
Reg Developer: OK, play the governance card, everybody else does (although you make a good point). But how accepted is data modelling in the community? How often do you need to explain the benefits to customers and potential customers?
I'd say that in the data community, having a data model is a fairly accepted principle, and it is rare that we need to explain the benefits (but it does still happen from time-to-time). Among the developer community, including some DBAs, buy-in is less accepted. I'd say about 50 per cent. I think the attitude and culture is different, more of the "I have a deadline to meet, I don't have time for design", "leave me alone and let me code", etc.
I think we can sum up by saying that understanding data is still important and there is still some point in the old disciplines of data analysis, even though we have moved forward in so many ways (business-centric development, XML and so on). In the end, however, we still have to deal with the old problem of development silos – business analysts, coders, database designers, and so on, all doing their own thing and not talking to each other. ®