Saturday, December 08, 2007

The Next Phase of SOA

I've spent the last couple days at the Gartner Application Development and Integration Conference in Las Vegas. As you might guess, the majority of the show was really about Service Oriented Architecture.

Like many SOA events, there were plenty of sessions on SOA governance, strategy, quality and implementation technologies. New this year was a strong focus on Web 2.0 as the consumers of the services.

Despite all of the good sessions, there was a void around "Information Modeling for SOA". The process of analyzing and designing an information model that mimics the business is a huge issue at every client that I work with. Many of our clients have moved past the initial planning and setup phases of SOA and are out designing and building services. Companies are realizing that having pockets of service teams (working in silos) leads to a collision of service semantics.

Nick Gall, Gartner Guru, commented that this year everyone seemed to be on the same page. The vendors, consultants, analysts and buyers were all in agreement about the steps you take to roll out SOA, the kind of infrastructure you need, etc. In essence, we've been doing it long enough now that there are well known success patterns that are rarely disputed. Nick was also quick to point out that we're not home yet. Like me, Nick feels that Information Modeling for SOA remains a deadly issue for many organizations. Whew! I was so glad to hear that I wasn't alone out there.

He drew the analogy to the database world. Today, most I.T. savvy people know the basics about how to design databases. However, it is a much smaller number of people who are experts in database design - the kind of people who can tell you the difference between 4th normal form and Boyce-Codd normal form. In SOA, we lack the equivalent set of rules for modeling services. In SOA, most people don't even know the equivalent of 1st normal form. Guys like Thomas Erl have thrown out the basics of service design, but it is hardly enough to avoid the pitfalls.

And the issue is much larger than just designing a single loosely coupled service. The real issue is designing a portfolio of services without creating redundancies or conflicts in the message model. Imagine taking the message models of all of the services across a domain (or set of domains) and reconciling their semantics. This view is takes the service out of the 'semantic silo' and begins to introduce an enterprise vocabulary. And like all data, relationships will exist. Here is where it gets interesting. Many people have focused on 'decoupling clients and services' (which is great), but have completely forgotten about decoupling the data model (or service model if you prefer). Pull out a large ER-model and look at all of the relationships between tables. When you resurface the data models as business or data services those relationships will still exist.

The collection of services and their message models is what we call the Enterprise Service Model (ESM). The science of creating the ESM is centered around "Information Modeling for SOA". More to come on this...

No comments: