Hi Bert, On Wed, 25 Aug 2010 14:44:31 -0500 "Manfredi, Albert E" <albert.e.manfr...@boeing.com> wrote:
> Mark Smith: > > > Possibly it will be surprising to a number of people on this list, but > > some of the ideas in IPv6 are over 30 years old, such as single, fixed > > size network and node portions, and using link layer > > addresses as layer > > 3 node addresses - > > > > "Address Mappings", Jonathan B. Postel, 2 May 1979 > > http://www.ietf.org/rfc/ien/ien91.txt > > Doesn't surprise me at all. > > I take it, your position is that if an address can have a 64-bit prefix and a > 64-bit host ID, then all you should need or want is classful addressing. And > only one class at that. Does that sum it up? > When it is possible, it is simpler to have a single sized node address, which also means the same sized prefix length. Simplicity benefits everybody - protocol developers, implementers, operators and end-users. In IPv4 it isn't possible anymore, and hasn't been since around 1981/82 (i.e. somewhere between rfc760 and rfc791), because there weren't enough addressing bits. Classes were a work around, and were the first of the "elegant hacks" used to overcome IPv4's limited address size as the scale of it's deployment started to exceed the original design assumptions. Fixed length subnets, then variable length subnets, then classless inter-domain routing were subsequent elegant hacks to get around these limitations.* With IPv6 we can have that simplicity back, because the address size constraint has been removed. We don't need to redeploy the complexities of IPv4 that we've become comfortable with - we've only become comfortable with them because we didn't have a choice. We're used to paying the price of that complexity, and don't notice it, because it was an essential cost of growing and operating the Internet. (* if you defined a protocol version change as occurring when the way packets are processed is changed at a fundamental level, e.g. address lookups, and software/firmware or hardware replacement needs to take place to facilitate that change, then each of these "elegant hacks" could have been a new version of IP - IPv6 might actually be IPv10 (allowing for IPv5 / Internet Stream Protocol)) > What's old is new again. We could almost revert back to RIPv1. > > I don't necessarily disagree, and I got that same message from others on the > list in the past. And I also wouldn't suggest that prefixes shorter than 64 > should be in heavy demand (except in assigning address blocks). It's just > that the restriction seems unnecessary, especially given that /127s and /128s > are already in use, that CIDR has been invented and everyone knows how it > works, and also because the /64 restriction only matters for SLAAC. > I think IPv6 "CIDR" i.e. longest match rule across the whole 128 bits is really only insurance against having to perform a whole of Internet upgrade, similar to what had to happen when CIDR was introduced, should the chosen boundary between the network and interface id portions end up being incorrect. Its likely it won't be necessary to use, as all the address sizing calculations, e.g. RFC3177, RIR policies etc., are based on the assumption of 64 bit interface ids. I think that is a wise decision, however it is only insurance - something you use if the unexpected occurs. Regards, Mark. -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------