On Oct 14, 2014, at 5:14 PM, James Woodyatt <j...@nestlabs.com> wrote: > But there is a problem with only deprecating prefixes without expiring them. > If they never expire, then they accumulate without limit within existing > networks as they join with newly commissioned networks over the course of > their lifetimes.
Ah, sorry, I didn't mean to say that we deprecate them but don't ever get rid of them. I think once a deprecated ULA has expired, it should be gc'd. If the homenet is partitioned, the two options are for the partitions to continue using one ULA and try to keep prefixes stable, in anticipation of the partition being healed later, or for both partitions to switch to new ULAs, or for one homenet router to "own" the ULA and get to keep it for use in whichever partition it winds up in, while the other partition has to choose a new ULA. Personally I think keeping the ULA stable across partitions is preferable, but I'm not sure it's possible to do it without the risk of flash renumbering. > So what's the problem? My language above ensures that home network hosts > always have at least one gracefully renumbered IPv6 address routable > throughout the entire network. If we need a further guarantee that hosts > always have an *invariant* address— which is an objective you've said above > that you think we don't actually have— only then are we faced with the > problem of prefix accumulation through network joins, which is a problem I'm > not sure we know how to solve effectively. My proposal avoids that trouble. I understood your language to be trying to get rid of all ULAs if any GUAs are present. Did I misunderstand? _______________________________________________ homenet mailing list homenet@ietf.org https://www.ietf.org/mailman/listinfo/homenet