[6bone] semi-newbie Q on IPv6 address planning

[Robert Elz]

    Date:        Fri, 2 Aug 2002 09:40:10 +0200
    From:        Gert Doering <gert@space.net>
    Message-ID:  <20020802094010.L27015@Space.Net>

  | If /127s are not used, what is the benefit of a /112?  Who needs 64k 
  | addresses on a ptp link?

No-one, but when written textually, /112 starts looking attractive to
use, and the other 48 bite between /64 and /112 are plenty for any site
to use to number its different p2p's.

The "we need 128 anycast addresses" argument that Pekka made isn't really
very important for this, on a p2p we don't actually need any.  But /112 is
nice to use anyway, and if it keeps people believing that those anycast
addresses are there available, just in case they're needed, then fine...

  | EUI-64 sucks big time (using MAC addresses and such is a nice idea,
  | but why on earth can't they map 48 bits to 64 in a somewhat more
  | straightforward way than "distribute them over all the 64bits"?) 

As someone else pointed out, it is the way that IEEE defined the mapping
for ease of allocation purposes.   I have the impression that we actually
had it right when it was first proposed, when FFFF was inserted instead
of FFFE, when I read the IEEE stuff about this, but it is way too late
to worry about that now.

| The existance of non-useful standards in other areas doesn't mean one
| has to use them for everything else needing 64-bit identifiers. 

No, in geneneral, you're right.

However, here the risk is, as these are all IEEE assigned numbers, that
some new net technology using 64 bit addresses will be developed, and
then some manufacturer will build an ethernet to X bridge/switch, which
converts the addresses on the fly according to the IEEE rules.   Things
would break big time if we'd invented some other way of doing the same
conversion in that scenario.   So, as it is all just bits, and the
mapping doesn't actually matter (we most certainly don't want "control
information" in the middle of addresses), there's nothing to be lost
by doing it the IEEE way (for which there are good reasons in their world)
other than the cost of explaining the reasons over and over again.

kre