Sunday, May 18, 2008

The Art of the Custom Documentum Object Model

Before designing a Documentum Object Model, you’ll need to take a litmus test of the culture of trust between the sponsor of the project and the IT organization. If the company is large there will be multiple levels of politics. You’ll need to judge from how the requirements gathering sessions went to figure out your approach to the object model design. Part of gathering requirements is educating your client on the types of objects that make up the content management system without getting too technical and wrapped up in explanations that are too long winded and lost on the client. During this education, ask questions like:

  • How do the different business units communicate with each other?
  • Do they share information, is there emphasis on security?
  • Are there databases that they use to look up information?
  • How effective are Marketing and Sales at driving the accumulation of knowledge into content published to consumers?

The answers to these types of questions help determine the meaning of the object model’s hierarchy levels. As an architect of the content management system, you are the only one who is qualified to make non-biased design decisions and hopefully would not have an agenda in your design. The most common object model hierarchy has an enterprise object as a child of dm_document, dm_folder, etc. and then children objects underneath it. The ramifications of your design magnify at the second level of the hierarchy. Here are some scenarios of what happens when the architect gets influenced in the wrong ways:

Forced to design without all the requirements
Have you been given enough time to really design a model that reflects the whole organization? If the IT Manager on the project says, “Don’t worry about the whole organization, we have three business units in front of us now, this is all we need to worry about for now,” you know there will be issues with the design if you create a model without knowing the bigger picture. Scalability, performance and reporting all suffer when a design is not drawn from the full foundational background.

Influenced by the wrong folks in IT
Most notorious for screwing up an object model would be the turf war database architects who don’t understand object-oriented design. They demand to know what the relationships are between all of these tables, where’s the schema, etc. They will flip out over table unions and joins if you tell them too much. They could care less how much the object model is integrated with the UIs and security. So when you say that the enterprise level attributes are reserved for only the most far reaching attributes across the whole company like retention period, stick to your guns when they push back and shake their heads. If you need more than 3 levels of custom object levels, make your case is as sound and simple as possible, try to include monetary impacts on potential customizations if it isn’t followed.

Confusing Department Security with Content Functionality
Most companies are set up by departments. Each department shares some information and restricts access to the rest. This doesn’t mean that the object model has to follow the org chart of the company. It may make more sense in the long run to figure out the function of each content type in the enterprise and really study the use cases of the content that is most critical to the company’s success. For example, for a government organization which has vital records (like birth certificates) to scan, index, and store, it makes sense to design an object model around function, in this case vital records, instead of the name of the agency that keeps track of the vital records. Agencies and departments will reorganize over time, functions such as birth records will not.

One repository vs. many
In many cases multiple repositories designed around one global registry makes sense. There’s more flexibility built into this design through out the technology stack, as well as with the changes in the business units over time. This however does not mean that each repository should have autonomy in its object model design. In fact, there should still be an enterprise level custom object for each object type being customized. The object model should be the same in each repository. You’ll have to be more diligent with migrating docapp archives between repositories, especially with the install options.

Decoupling Internal Business Process and External Publishing
If the end goal of the content is to publish it to a portal, there will be conflicts between the internal structure of the content (how the business works with each other) and external structure (how consumers view and search the content). Do not underestimate how long it will take to weed out the navigational systems for each side, the security and identity management, the functional driving attributes, etc. In the best case there will be enough decoupling of the objects and their attributes that the design can provide decoupled and scalable solutions to the conflicts between content management and content publishing.

Some General Rules of Object Model Design

  • Determine if content types are functional or departmental in nature
  • The security model of the repository whether it’s user, object, or folder based may have an overriding influence
  • Build in flexibility to enable the object model to expand in all directions
  • Move attributes that span all of object types up one level if possible