On 13-jul-2006, at 12:21, Iljitsch van Beijnum wrote:
draft-narten-ipv6-3177bis-48boundary-02.txt
After speaking with Thomas shortly in the hallway yesterday I understand that he doesn't feel it's appropriate (anymore) to tell the RIRs what prefix sizes to use. I don't necessarily agree with that, but in the interest of moving forward I suggest that we try to talk about what the IETF is supposed to know about: technical issues. When we agree on those the rest should be easy. :-)
Since the assumption until now has been that pretty much everyone would receive a /48, it makes sense to determine whether this is something that the addressing technology as we understand it today can support. In my opinion, that answer is:
Assuming the 2000::/3 unicast address space and a reasonable HD ratio target of 80%, it should be possible to support 2^(80% * 45) = 2^36 = 68 billion /48s. Based on this, there is no technical reason to deny anyone a /48.
Then there are technical considerations for a minimum size assignment. Today, most end-users connect one or more hosts in their site to the internet through an intermediate device that, amongst other functionality, functions as an IPv4 router. It's reasonable to assume that this situation translates into a scenario in IPv6 where users have a router on their site. This requires at least two subnets: one internal to the site, that hosts connect to, and one used between the user's and ISP's routers. However, there is no requirement that these two subnet prefixes (which should be /64 as per relevant specifications such as RFC 3513) come from a single shorter prefix assigned to the user. This means that the minimum for most users would be a /63 or two /64s.
Next: preferred prefix sizes. DNS reverse delegation happens on 4-bit boundaries. It's not easy to draw conclusions from current operational experience with the reverse DNS in IPv6, but it's not unreasonable to assume that more than with IPv4, with IPv6 users will run their own reverse DNS servers. This gives a slight preference towards assignments on 4-bit boundaries.
Finally: stable prefix size when changing ISPs. Both for ISPs and end- users it's easier if there is only a small number of prefix sizes in use. This avoids the situation where a user would receive a smaller assignment from a new ISP, making renumbering significantly harder, and for ISPs, simplifies reusing address space reclaimed from users that have left.