6bone Architectural Changes? [RE: pTLA request for RMNET - review closes 23 April 2002]

[Pekka Savola]

On Wed, 10 Apr 2002, Michel Py wrote:
> >> Bob Fink wrote:
> >> If you think there is something wrong with this, please say more.
> 
> > Pekka Savola wrote:
> > If I was serious about experimenting with IPv6, I sure would not
> > go overseas to find someone who might be willing to slice off a
> > part of a block.  However, if I was not serious, I wouldn't care if
> > all my IPv6 traffic to European countries crossed the Atlantic twice.
> 
> I have to disagree with you on this point. I am in California, and the
> first tunnel I got was from Viagénie in Québec. Yes, I could have asked
> Hurricane Electric, or I could have asked Bob Fink, or I could have
> asked Cisco.

I think an ISP and a user asking for a block are two different things.  In 
any case RTT from Viagenie from Europe is around 250ms apparently.

> Who cares if my average RTT to Viagénie is 92ms? Not me. There are ways
> to adjust network latency in Quake v6 (the only IPv6 real application as
> of today, I think?)

Sometimes IPv6 is actually used in e.g. SSH sessions, WWW-pages etc.

Nobody will want to do stuff over IPv6 if the quality is always low.

> > In real use this just would *not* have been acceptable.
> 
> The 6bone is _not_ real world. There is no real world, as of today. When
> we get native links, it will be time to re-assess the situation. (Bob, I
> can provide you with a repeater in Sacramento when you lay out that dark
> fiber between Berkeley and Truckee ;-)

Native links are a sufficient but not required condition.  It would be
sufficient that tunnels are only done locally "within a country or a
region", or *at least* so that people with long tunnels don't use them for
transit without thinking REALLY HARD about it.

> As far as I am concerned, the geographical location of a pTLA does not
> matter (I have a tunnel with UK, too). There are situations where the
> link to someone geographically close will be worse than someone at the
> other end of a continent. To some extent, I think that cross-ocean
> tunnels are a guarantee that the 6bone will not be used for production.

Let me modify your statement above a bit:

"I think the tunnels are becoming a guarantee that IPv6 will not be used
for production."  [Note: I'm not saying all tunnels are bad, quite the 
contrary]

This is still true if one assumes a goal of IPv6 is to be deployed on a 
global scale.  Some might disagree about the timing and the scope.

The only ways I could see to get rid of the "cruft" of 6bone would appear 
to be like:

 1) Disband 6bone.  (This may not help in a global scale anyway.  But
how??)

 2) RFC2772 policy changes that "transit" MUST NOT be provided over
non-local tunnels unless there are strong reasons to do so (and enumerate
some reasons).

 3) People who are serious or semi-serious about IPv6 create a blacklist
of certain 6bone AS's: AS-PATHs which contain these AS's as non-terminal
members are rejected.  This blacklist would include all 6bone "toy"
transits.  This would kill legal traffic too, though.. A drawback here is
that everyone these serious IPv6 people want to connect to should use
similar policy so that return traffic should not be "blackholed".

 4) People who are serious about IPv6 refuse to talk to any 6bone
addresses, use the blacklist above, and an additional blacklist of AS's
which have only 6bone addresses. (Or a whitelist..). This is possibly
sufficient to keep off the "6bone pollution".

 5) Others?

I think we should look at option 2) or 5).

I think this is an "architectural discussion" that might have been useful
to have been included in 6bone meeting in IETF53.  However, it might have
just degenerated in a flamewar so perhaps this is for the best.

In any case so that 6bone does not become dead weight that tries to pull
IPv6 under the surface, we need to think of ways how to make IPv6
connectivity work better than now.

-- 
Pekka Savola                 "Tell me of difficulties surmounted,
Netcore Oy                   not those you stumble over and fall"
Systems. Networks. Security.  -- Robert Jordan: A Crown of Swords