[rrg] Summary of architectural solution space

[Teco Boot]

Hi Bill,

/ RFC 1887 4.4.2 is obviously not functional with *currently deployed
/ network stacks*. It would need the kind of software described in
/ strategy B or H in order to be viable. Or do you disagree?

It depends.

This solution needs SA routing towards DFZ (like my BRDP Based Routing
proposal). Maybe lack of easy_to_manage_PBR / automatic PBR is the main
reason that this option is not used that much. Easy to fix!!

As mentioned earlier, I experience renumbering each day. And I *am*
reachable. So examples of "kind of software" is available and massively
deployed.
www.ietf.org / www.irtf.org are accessible using PA addresses:-). Hosted
servers can provide high availability. And it is not that difficult to make
servers PA multi-homing. I see two concerns:
 1: Support for changing ISP (multi-homing with slow renumbering)
 2: Support for session continuity when link to ISP fails 
Only 2 is hard to fulfill without upgrading host stacks. But I doubt if many
of us need this. 

I do not think mobility should be handled by the routing system. We have
protocols for this (MIP6).

So I would say: some of us need an upgraded network stack. We have already
two solutions (HIP & Shim6) (use your sshagentd for this?).
Problem here is, solutions need updated stack on corresponding node.


/ > On Strategy B and H:
/ > I think the difference of the two strategies is minor. Make methods
/ A2a and
/ > A2b for these?
/ 
/ I vacillated on that for a while and eventually came down on the side
/ of calling it a different strategy. I think if the difference was
/ truly minor, IPv6 would have worked out a whole lot better than it
/ has.

Maybe this can be corrected. 
Do you really think the Internet would work a whole lot better if we go for
this strategy?
I am with you!!


/ > And on the criticisms:
/ > With B: Why would LOCs be constantly in flux? Fixing this should be
/ part of
/ > the strategy.
/ 
/ Because link state changes on the nearby upstream path are satisfied
/ by renumbering instead of rerouting. When I said "dynamic," I really
/ really meant it.
/ 
/ Actually, that's not strictly true. You'd reroute for a few minutes,
/ long enough to be sure the state change wasn't ephemeral. Then you'd
/ renumber and on completion discontinue the exception route. That's
/ because renumbering looks like a distance-vector protocol; I can't
/ think of a way to describe it as a faster link-state protocol.

DV can distribute triggered updates quite fast. BRDP supports this. In some
cases, DV is faster than LS.

When a link is flapping, why using the prefix that comes along with it? We
need metrics & hysteresis for address assignment and SA selection, just as
we have in routing. All part of BRDP (better described in next versions). In
a MANET, this is mandatory for obvious reasons.


Teco.