On Tue, 05 May 2009 13:04:49 +1000 Karl Auer <kauer@biplane.com.au> wrote:
On Tue, 2009-05-05 at 04:49 +0200, Randy Bush wrote:
I'm with you. I wish vendors and spec designers would just get over it and let people subnet however they want. [...] do other than 64 and you do not get auto-conf. some do not consider this a loss, others do.
This is an important distinction.
- you CAN subnet however you like, with any number of bits in your prefixes
- autoconfiguration will work only in subnets with a 64 bit prefix.
The two matters are quite independent of each other, as far as I can tell.
Older protocols, like classful IPv4, Appletalk etc. put a hard boundary between the network and node portion. That was simple and, in the case of IPX, Appletalk and DECNET, it was very convenient to have fixed length network and node portions. IPv4 originally had a single boundary between the network and node portion - if you look up the early RFCs/IENs, the IPv4 addressing format was similar to Class A. Of course in the case of IPv4, those classful hard boundaries caused problems when we needed to squeeze more addresses out of the 32 bits by moving to a fully varying boundary between the network and node portions. IPv4 software in all nodes needed to be upgraded to work. I think of the way IPv6 has done it is the middle ground. For forwarding, the boundary between the network and node portions isn't hard - it's purely longest match on the whole 128 bits. However, because we've got so many bits, within a portion of the address space, a harder (but not hard) boundary exists, to benefit from the convenience of having fixed length node addresses, which results in things such as much simpler autoconfiguration etc. Regards, Mark.