> I was thinking the same. Above model would just work fine if we want to > use libscf.so/svc.configd to just store the interface configurations and > not represent any of the IP interfaces as SMF instance(s). > > > > Certainly, moving the configuration to SCF doesn't require representing > > individual IP interfaces as service instances > > but is it OK, to just have one aspect of SMF (configuration part) and > not have any instances of the service which represent the configuration > i. e. use SMF's configuration framework and not use SMF's service > management (start, stop, refresh) framework.
FWIW, the SMF team (and Shapiro in particular) were extremely opposed to using SMF as a "dumping ground" (his words, IIRC) for networking state. There is a lot of history here. As I mentioned earlier in this thread, there's also a lot that the Fishworks team has learned operationally from implementing their interfaces-as-instances model. How serious are you about going in this direction? -- meem
