At 04:43 PM 3/13/96 -0800, Michael Dillon wrote:

Here's an idea. Let new ISP's reserve large blocks (say /16's) in 65/8, 66/8, .... but don't let them actually use these addresses on the global Internet. Then, the ISP can run a Network Address Translation gateway and give their customers 65/8 addresses while still using a chunk from their provider's block. And they can switch providers without forcing their customers to renumber. Then, after they have demonstrated that they should be given a /16, open up the block they were given in 65/8 for use without the NAT.

Well, I knew this topic would come up sooner or later, since it was discussed briefly in LA/IETF at the CIDRD WG meeting(s). A snippet from the LA/IETF CIDRD minutes: [snip] Yakov Class A Allocation Guidelines ============================= Motivation Observation 1: 192/3 is 1/8 of the total IPv4 unicast space Observation 2: 64/2 is 1/4 of total IPv4 unicast addrses space Hypothesis1: at some point we will exhaust 192/3 Hypothesis2: at some point we will need to allocate out of 64/2 Observation3: It appears safe to allocate out of 64/2 based on experience documented in RFC???? on Class A Experiment Recommendation allocation of /17 or larger should be done out of 64/2 allocation of /18 or smaller should be done out of 192/3 Comments? Randy: Do you mean start this now? YR: No, only when we decide it necessary. Eliot Lear: Can big ISPs get bigger blocks? YR: Yes, it should allow them to. EL: Do we need to advise Registries? EL: Should 192/3 be declared unroutable? YR: Perhaps part of it. BManning: I would like to attach a rider on this. I think before anyone gets anything out of 64/2 that they should agree to give back their old blocks. TonyLi: Is there some reason not to start alloc 64/2 immediately? YR: We don't have to yet, so perhaps we should not. TonyLi: Pressure from international carriers to get large blocks. ELear: We should wait as long as possible to age legacy systems. Tony: There weren't any big problems in the RFC. BManning: Some things were not tested. Only routing protocols were tested. We couldn't figure out a reliable way to test interior routing. We should hold off a little longer before we jump into this. NoelChiappa: We can't get rid of all the legacy systems. Is the cost of hurting the legacy equipment less than the benefit of 64/2. [general discussion of when we should implement this. Now or wait a little longer for legacy systems to expire.] [discussion of route-able v unroute-able prefixes. Will certain prefix ranges be declared unroute-able in future?] MKosters: @home has a /14 out of net 24, so we have already allocated out of Class A. BrianC: Suppose IANA decides to do this, and some joe-ISP asks for a /14. We haven't given the Registries any guidance on how to allocate. RConrad: Policy most likely to remain the "power of 2" increase. The size of the ISP is irrelevant with this scheme. Cathy of @home complained that Sean won't take 24/14 only 24/8. Sean Doran raised the issue of charging for prefixes. Tony advised we leave this for further discussion on the mailing list. [snip] - paul