Recently, a client approached me with a quandary.  When designing XML schemas for Web services, how do you balance the desire to use industry standards such as UBL ( Universal Business Language) or CICA ( Context Inspired Component Architecture) to support data interoperability with the unique needs of particular domains and sub-systems within the enterprise?  The client’s service design team is rightfully concerned with the competing interests of internal enterprise standardization, interoperability with external entities, and addressing the unique needs of local domains and process constraints. How can these competing needs be effectively managed when designing the schema for a given service?

The Standard Answer

Naturally, you start by giving the standard answer - “it depends”.  This is an essential and carefully crafted phrase that all consultants are taught to give according to page 17 of How to Win Clients and Influence Budgets. Of course, the standard answer is rarely sufficient, but it is not without merit.  It is certainly true that the decision regarding your data model design depends significantly upon a host of factors:

• Which use cases are high priority?
• Which use cases are high impact (to stakeholders and/or to customers)?
• Which systems and/or processes are mission critical?
• Are you including industry standard data models in you enterprise out of necessity (i.e. ONLY when you are forced to interact with other entities), or is it part of a broader modernization effort within your organization?
• Do interoperability challenges or integration challenges currently contribute most significantly to your development and/or maintenance costs?
• Which interaction scenarios are of higher priority, internal or external service calls?
• Would aligning more closely to one of the available models lead to a significant reduction in data mapping activities or is the environment simply too fragmented to support this?
• Is there one particular system or business process that is unique and skewing perceptions regarding our service data models?

While all of these questions (and many more) are important and aid in facilitating a thorough examination of the problem, we rarely have the time to examine each problem from all possible angles.  Consequently, we must look toward guidelines and rules of thumbs.

Guidelines for Managing Multiple Service Schema

1)       Your business processes should only work with a single data model if at all possible.  Business processes are best designed to be data model agnostic and operate off of an internal, process-centric model.  I explained the importance of process-centric data models in a previous post.

2)       Your services should work with as few data models as possible (one model being preferable).  This is just good commonsense.  For each data model that is added to the mix, your development time and long-term maintenance costs increase at a non-linear rate.  The pain will intensify and it will do so rapidly with each new model you add to the mix.

3)       If your services are going to work with multiple models, you should put some sort of taxonomy / categorization scheme in place to distinguish the data models used by services.  For example, services that are outward facing might use a data model accepted more broadly by the industry.  Services used by a particular LOB might use a certain data model, and services used by another LOB might use another.  Another distinction could be infrastructure services vs data services vs application services.  Regardless of the approach, there needs to be some methodology that is objective and governable for when one data model is used vs when another data model is used.

4)       Data model transformation is a necessary evil.  It should be done only when necessary and you should contain the transformation to a designated component.  Transformation activities should be handled by intermediaries (data services, ESB, network appliance, other mediation framework) when possible rather than building it into the internals of a service or process.  This keeps your services clean, provides a nice reusable transformation mechanism, keeps your interoperability more loosely-coupled, and provides for agility and extensibility in the future.

Juggling multiple data models within a service oriented environment is no one’s idea of fun.  When possible, aim for a more comprehensive and strategic analysis of the environment (see the ‘Standard Answer’ outlined above).  When this is not realistic,  try to use the above guidelines and rules of thumb to help you tactically navigate the murky waters of data model incongruity.  Service design isn’t easy, but it doesn’t have to be rocket surgery either.  Best of luck!