Thanks for the comment David. I think we’re in agreement, I just didn’t express my thought clearly. At the model level (forgetting serialization), I think we need to have a wide graph and RDF is the right way to model it. Then, within each node, we don’t necessarily need to push the graph model all the way down to the description of all aspects of the node. For that it’s fine, for example, to use an XML format that is in no way related to RDF. In other words, use RDF (with any serialization you want, XML or other) to implement a wide graph and then use specialized model/serialization (which again could take the form of XML) within the detailed records.
That’s what we did for CMDBf. There is a graph model (alas, not RDF) linking items but the records of each item are whatever XML you want. So we use a graph query language (specific to CMDBf) coupled with XPath queries at the level of individual records. That’s what I was trying to express. No need to force a graph model on the entire data space, it’s good enough to have an overarching graph model and then go down to XPath or something else within the node. My point wasn’t about serializations.
Not that I’m trying to talk anyone out of going with a graph model all the way down if that works for their task, of course.
