> On 23.7.2015, at 10.49, Markus Stenberg <markus.stenb...@iki.fi> wrote: > >> On 23.7.2015, at 10.41, Juliusz Chroboczek <j...@pps.univ-paris-diderot.fr> >> wrote: >> Right now, the interaction between the routing protocol and the rest of >> the stack is very simple and very clean: HNCP redistributes assigned >> prefixes into the RP, and the RP redistributes the default route into the >> RA server. Redistribution is a well understood, widely implemented >> mechanism, one that we all feel comfortable with. > > To be more specific, > > [1] HNCP running daemon (e.g. hnetd) _configures interfaces_ which causes > local on-link routes to show up in the local RIB, and eventually get grabbed > by the RP. > > [2] RP redistributes them to other nodes' RIBs. > > [3] RA server grabs the route from RIB and everyone wins. > > I prefer this sort of loose binding, and bunch of other protocols deal with > local RIB too, so I believe HNCP 'fits' the typical model best this way. > > +- the SHOULD border discovery setting of allowed routing protocol > interfaces; obviously, we could just deal with _firewall_ and not touch RP at > all. I believe we should ensure that HNCP spec only states which interfaces > should not have router-router configuration _such as routing protocol_, and > leave actual HNCP implemenation - RP interaction implementation choice. > (Firewall, reconfigure, rewrite config + boot in the head, ..)
Baed on rereading the text, I am not sure if that was actually helpful; in case of e.g. Linux, there isn't really FIB-RIB separation and essentially you have only FIB (what is shown by e.g. 'ip -6 route'). Anyway, s/RIB/FIB/ if the description sounds odd to you. -Markus _______________________________________________ homenet mailing list homenet@ietf.org https://www.ietf.org/mailman/listinfo/homenet
