At 9:28 AM +0100 2/7/02, Gert Doering wrote:
different from IPv4, "renumbering [in IPv6] is easy" (supposedly)...
No, no, no, no, no. IPv6 has some features to make renumbering easiER, but it still may not be easy for some or even many sites, due to the many ways and places in which IP addresses are used (for most of which we do *not* have automated means of renumbering). Please do not adopt policies that assumes renumbering will be entirely painless for end customers, let alone for ISPs. Among the conflicting demands of conservation, aggregation, availability, etc. should be included "stability", where a "stable" address assignment is not a permanent- for-life assignment, but rather one that ought to be good for at least a couple of years in most cases, and one that ought not to be changed for less-than-compelling reasons. And on the general topic of discussion, I agree with Thomas. Unless and until we figure out how to do globally flat routing, it is vital to ensure aggregatability to a reasonable number of prefixes. I understand and appreciate the difficulty of defining "who's an ISP", and especially "who's a top-level ISP"; the RIR's "slow start" approach seems like the least bad scheme that anyone has yet been able to come up with, despite its flaws, so I'm surprised to see it discarded -- for IPv6 (if that's what's happening). I do wish we could converge on one of the possible multihoming solutions that would reduce the pressure for multihomed sites to seek PI prefixes... As for the research nets serving 50 universities, I don't think they should get /32s. If they were instead given longer prefixes and if, at some point in the future we end up with so many such longer prefixes that ISPs are forced to filter some of them out, the (presumably) relatively small number of especially "worthy" long-prefixes could be granted exceptions in the filter rules without too much operational effort (the real effort being the political one of identifying the worthies). The HD Ratio calculations that have been used to justify the current address field partitioning assume that the high-order /48 is being carefully managed, not squandered, so non-extravagance is still required at that level (i.e., something more generous than "conservation" but less generous than just giving huge blocks of /48s to anyone who asks). Steve