At 23:33 +0300 9/19/02, Pekka Savola wrote:
I'd perhaps reword that as "no global tunneled infrastructure". And even
then I'd be a bit hesistant: tunneling across short distances is not that
big a deal, but adding different administrative parties can make it more
and more difficult.
Solutions like ISATAP can provide rather good performance in a limited
case; connections outside of the "site" could be native v6.
That's a good point. I'm thinking of the global IPv6 infrastructure and
individual hosts or sites connecting to it. And I suspect that any
solution for this case that uses tunnels, automatic or manual, is likely
to face the problem of sub-optimal forward and/or reverse paths that you
mentioned originally. In other words 6to4 is neither better nor worse,
just different.
automatic tunnels techniques usually mean (by design) that the path is not
"controlled".
As long as the density of tunnels is much less than the density of the
connections in the underlying network, the path taken by the tunneled
traffic between two points will be likely to be worse than the path taken
by native traffic. And the well-known scaling problems of tunnels
(including crossing administrative domains) makes it unlikely that they
will become very dense without suffering maintainability problems.
Since there are relatively few 6to4 relay routers out there, performance
is often poor for 6to4 to native traffic. But the same is often true for
tunnel brokers, and sometimes for manual tunnels. And at least 6to4 has
good performance to other 6to4 sites.
semi-automatic tunnels techniques like tunnel brokers can be deployed so
that it does have less sub-optimal forward/reverse paths. It depends on
where/how you deploy them. If they are near to you, network topologically
speaking, then the sub-optimal paths might be not that a problem.