The article says that this is an "unusual architectural choice". I strongly disagree. A reference is made to an RTI publication many years ago. Realtime Innovations have been key players in the development of DDS publish-subscribe middleware for donkey's years. I remember NDDS back in 1998. We've successfully used DDS in the military space to enable different suppliers to connect to a DDS backbone-based system via the 'data model'. Especially in DIL environments (Disadvantaged, Intermittent, Lossy) where QoS settings can be applied to different pieces of the data model over a plethora of different bearers (think Satcom, IP radios, etc). It's a very-much typical architectural pattern selected.