On Oct 17, 2014, at 10:54 AM, Michael Richardson <[email protected]> wrote:
> I will go back and read James' message about joins and splits.
> It seems that we have this problem with GUAs as well, and it seems that
> the whole address selection issue exists without ULAs, as long as one has
> multiple ISPs.

The issue with joins and splits is that you would like your numbering to 
survive a temporary network partition, but to not survive a permanent network 
partition.   I think this problem is solvable, and have some ideas about how to 
do it, which are half-baked and probably won't work without tweaking.

E.g., when a homenet chooses a ULA, it could divide the /64s evenly between all 
of the participating homenet routers.   If there's a partition, the homenet 
routers get to keep the /64s they got to begin with, but if after some period 
of time (TBD) the partition hasn't healed, one of them (chosen back when the 
network wasn't partitioned) begins a transition to a new randomly-chosen ULA.   
If new routers are added to a homenet, HNCP should be able to balance out the 
pool of free /64s so that every router has some to keep during a partition.

If two homenets each with their own ULA are joined, one of the ULAs is 
deprecated with a long tail.   It would also be worth factoring power-cycling 
into this: if a homenet router that had a ULA is powered off, then powered back 
on, connected to a new homenet, its ULA is probably the one that should be 
deprecated.   This will still fail in the case that there are devices that are 
still using the prefix, but I think we're now down in the really unlikely 
scenarios that we don't have to solve.

_______________________________________________
homenet mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/homenet

Reply via email to