ietf
[Top] [All Lists]

IPv6 local experiment draft

2000-08-17 11:50:02
        not sure if it is fine to continue it here, i may want to move
        to ipngwg.... ietf(_at_)ietf looks too wide distribution.

| First of all, do not cook up IPv6 prefix on your own.  You cannot pick
| random prefix number, that can jeopadize the whole point of experiment.
Could you expand on that please?  So far, I see from Matt Crawford a single
reason (IP6.INT coordination), and that at least should be in the
draft, but I personally would be interested in what the objections
to a random selection of a global address really are.   An aesthetic
argument is fine, as long as it's an explanation of how it "jeapordizes
the whole point of experiment".

        i saw couple of incidents where novice user picked up:
        - multicast address prefix for unicast use (ff00::/8)
        - link-local address prefix for offlink use (fe80::/8)
        - non-standard prefix length (not /64, varies very much)
        and caused too much trouble for them.

        the goal of the draft is to help novice users from going into
        common pitfalls.

If it is possible, try to contact nearest upstream 6bone site, or
upstream ISP, to give you an IPv6 prefix
How does one evaluate the nearness of a 6bone site or ISP, so that
one can determine which is nearest?   What if the nearest is uncooperative?

        usually, 6bone sites connect to your location using IPv6 over IPv4
        tunnel (RFC1933).  for stable operation of the tunnel, it is very       
        important to make IPv4 hops very short.  if you configure a long-haul
        tunnels (like transpacific, or from eastcoast to westcoast across
        different ASes), you will have trouble with stability.

        use IPv4 traceroute to measure the nearness.

        if the nearest is uncooperative (i hope not...), pick the next closest
        one :-)

o The site may use the address prefix: 3ffe:0501:ffff::/48.  The address
 prefix was curved out from WIDE 6bone prefix.  The site MUST be
 renumbered, before the site gets connected to the worldwide IPv6
 network.
Shouldn't IPv6's much-trumpeted stateless autoconfiguration and
renumbering scheme take care of that, well, automatically?

        the phase-out happens only if the site adminitrator properly configurd
        for that.  we need to put enough warning as we will have major trouble
        if the address leaks out to IPv6 worldwide network (just like
        IPv4 private address leakage).

How is this different from a completely-disconnected site using
a randomly selected prefix?

        see above.

itojun