Jonathan Altman

Found this quote today from Eve Maler

The trend in distributed computing is towards service-oriented architectures (SOAs). Early in the life of this buzzword, some people said it should really be called a document-oriented architecture (except for the unpleasant acronym :-) because it’s really all about document-passing rather than a tightly coupled RPC paradigm, that is, if you want to be successful at reusing the components

The quote is part of a longer discussion that is partly tangential, but I have spent the last 4 years dealing with integration of services that are or could be called "web services" and they have taken various approaches to how the XML data is moved. Some have expected that we will use a SOAP toolkit that hides the XML behind objects that get serialized and de-serialized. Others came out of a more EDI-like world and are more message or data structure passing oriented.

The ones that are easier to deal with are the ones that move messages or data structures, not insist that I hide behind the de-serialized objects. In general the objects end up not serving my purposes well, and require large recompiles and test cycles to deal with. We end up having impedance mismatch between the software artifacts in the web services we consume, expressed in objects that I must use but which I did not define, and my own software artifacts. To fix this, I need to either write wedges that sit between the web service I consume and my software artifacts and do the mapping, or I need to build my software artifacts so that they have a "has-a" relationship with the web service's artifacts. That just splits the wedge into per-object pieces where each of my objects that "has-a" needs to manage its map. The really intractable problem is when the object mapping that one toolkit makes is incompatible with another's.

On the other hand, when someone changes an XML document that I can treat as an XML document, I can work in a script-based (not compiled) environment and update the mapping between the document and my software artifact the way I need to, and ship faster. I have variance between what you are sending me and what I expect, but since my entire system is built solely as a mapper between my artifacts and the web service's documents, there is no impedance mismatch, just a version change to my mapper.

It's faster to update, easier to maintain, and I don't have to care about whether your web service's underlying object model is any good or not. I have one system, the mapper between documents and my software artifacts, not 2 systems: your objects and my wrapping layer to handle the impedance mismatch. And I cannot end up in situations where my toolkit and yours cannot generate compatible object<->message bindings.

At the end of the day, I think you have to care about what's in the angle brackets. Developers who just want to deal with objects see a false economy: you feel like you are working higher up the protocol stack because you do not observe the wire format--just your objects. Eventually, however, you end up working below the wire format because the underlying plumbing of the service you are talking to is exposed in the software artifacts (objects) you end up having to manipulate on your side. You either have to know details about how the objects on the other side were built, or you have to muck with the document format anyway to map away details from the other side you don't care about or cannot handle, but usually both.