Thursday, May 05, 2005

Service Descriptions - There does not need to be just one

There has been some really good analysis and thinking about service description languages recently by several smart people (Tim Bray:SMEX-D, Norm Walsh:NSDL, Mark Nottingham). Everyone of them makes some excellent suggestions and brings a unique POV. One of the challenges that everyone is assuming is that there needs to be only one service description language. I challenge this assumption and suggest that more than one is actually preferably.
The are a few reasons for wishing there should only be one service description language:

  • We have a mind set that is framed by the Web publish-find-bind triangle where we assume that magic happens by everyone automatically discovering services and magically binding to them to solve complex business problems. This implies that there is a many to many relationship between services - all services need (and can) talk to all other services. In the real world this is more like one to a few or few to a few - a slight over design.

  • There is the burning desire for protocol independence - between companies there is unlikely to be a placement for HTTP/SSL anytime in the future. As the enterprise architecture is slowly being decomposed into services provided by outside organizations I would suggest having a single transport is more important than have transport/protocol independence. If the transport information is cleanly separated from the message there should be no issue in replacing it with a different transport - a good example is EDI and AS/2
  • The complexity of integration is not in the communication of messages it is in the integration of the semantics after the message has been received. Currently we are doing several integrations a month with a wide range of systems. The message delivery is not the problem it is the semantic meaning of the messages and this is in a single well defined vertical.

  • Tool development has always been a good reason for having a single set of standards that all vendors build. The tools are complex because the service description is complex, and there is a lot of complexity introduced by RPC. If we focus on the message (where the real integration occurs) there are really good tools available, XMLBEANS for Java does a truly great job of providing a binding between XML and Java, similarly XSD for .NET and I am sure there are many others. These tools work on basic schema's and are generic tools - rather than specific more complex tools.

One of the really good aspects of the service description languages described by Tim Bray, Norm Walsh is that they use the same vocabulary as the transport i.e. HTTP -this makes it a lot easier to understand as it does not add another layer of verbal indirection. This reduces the learning curve and hopefully the ambiguity. Having several service description languages is not an issue if they are all in xml they can be transformed from one to another easily. As the transformation is a design time dependency and not runtime a runtime one there is no performance penalty. The major issue may be loss of meaning but there are only a limited set of MEP's so this should not be a long term issue.

The tendency today for web services is to add more features and hence the complexity is increasing. Shift the focus to how simple they can be and reduce complexity, lets see how easy it can really be.

The focus must be on the communication of information and not overly abstracting the service descriptions to a point where everyone just gives up and writes word documents.

No comments: