There’s one huge problem with this conference: too many interesting sessions going on simultaneously, so I’m sure that I’m missing something good no matter which I pick.
I finished the morning breakout sessions with Woody Woods of SI International discussing transitioning enterprise architectures to service-oriented architectures. He started out defining SOA, using the RUP definition: a conceptual description of the structure of a system in terms of its components and the services that they provide without regard for the underlying implementation of the components, services and connections between components). There are a number of reasons for implementing SOA, starting with the trend towards object oriented analysis and design, and including the loosely coupled nature that allows for easy interfaces between systems and between enterprises. Services are defined by two main standards (in the US, anyway): NCOW-RM, the DoD standard, and the OASIS reference model for SOA.
There are a number of steps in defining operational activities
- Establish vision and mission
- Determine enterprise boundaries
- Identify enterprise use cases
- Detail enterprise use cases to create an activity diagram and sequence diagram
- Develop logical data model
The process for developing a service model, then, the following steps are taken (using RUP terminology):
- Identify the roles in a process.
- Identify the objects in a process, starting with triggers and results, and refining to include all objects, the initial actions and a complete action analysis, potentially creating a sequence diagram. Other types of process models could be used here instead, such as BPMN, although he didn’t mention that; they’re using Rational Rose so his talk is focused on RUP models.
- Identify boundary crossings, since every time an object crosses a boundary, it’s a potential service. By “boundary”, he means the boundaries between roles, that is, the lanes on a swimlane diagram; note that some boundary crossings can be ignored as artifacts of a two-dimensional modeling process, e.g., where an activity in lane 1 links to an activity in lane 3, the fact that lane 2 is in between them is irrelevant, and the boundary crossing is actually between lane 1 and 3.
- Identify potential services at each boundary crossing, which implies encapsulation of the functionality of that service within a role; the flip side of that is that it also implies a lack of visibility between the roles, although that’s inherent in object orientation. Each boundary crossing doesn’t necessarily form its own unique services, however; multiple boundary crossings may be combined into services (e.g., two different roles requesting information from a third role would use the same service, not two different services). In this sense, a service is not necessarily an automated or system service; it could be a business service based on a manual process.
- Identify interfaces. Once the potential services have been defined, those interfaces that occur between systems represent system interfaces, which can in turn be put into code. At this point, data interfaces can be defined between the two systems that specify the properties of the service.
In this context, he’s considering the RUP models to be the “enterprise architecture” that is being transitioned to a SOA, but this does provide a good methodology for working from a business process to the set of services that need to be developed in order to effectively implement the processes. I’ll be speaking on a similar topic — driving service definitions from business processes — next week at TUCON, and it was interesting to see how Woods is doing this using the RUP models.