Sam Hartman wrote:
"Keith" == Keith Moore <moore(_at_)network-heretics(_dot_)com> writes:
Keith> It seems to me that the general problem is not multiple
Keith> interfaces, but multiple addresses per host. It doesn't
Keith> matter (much) whether those addresses result from multiple
Keith> physical interfaces, a combination of physical and virtual
Keith> network interfaces, multiple prefixes being advertised on
Keith> the network attached to any particular interface, or even a
Keith> mixture of v4 and v6.
Keith> So that might have some impact on the name, particularly if
Keith> you want to attract the breadth of interests whom this
Keith> affects. Something like Hosts Addressed Multiply (HAM),
I'd actually appreciate focus on the multiple interfaces (or multiple
network providers) problem. I think that attacking this in full
generality is well beyond what we can manage. I think even a focused
problem may prove challenging.
I certainly agree that it's challenging to attack the generalized
version. However, if you try to solve each version of this problem
separately, the result will be even more complex, less workable, and
less realistic than if you try to look at it from a broader view.
There's a strong potential for those different solutions to collide with
Or to put it another way, if you define a solution to a narrow version
of the problem, it might not actually solve anything in the real world,
and the WG's work will have been wasted.
And from the POV of the application, or the transport layer, it really
doesn't matter much _why_ the host (or peer) has multiple addresses - it
still has to deal with them.
As for whether "attacking this in full generality is well beyond what we
can manage": Based on previous work that I have done in this area, I
suspect that this is exactly the case. Which is why I think that it
should be reasonable for the WG to look at the problem and come back and
say "we haven't identified a good solution to this problem, and the best
practices that we can recommend are to minimize the cases where a host
or app has to deal with multiple addresses for itself or its peers".
Of course, a better result would be for the WG to say "we recommend that
the use of multiple addresses per host be limited to these specific
cases, and avoided in other cases"... along with specific
recommendations for networks, stacks, APIs, applications, etc. I think
that this kind of output would be the most useful that the WG could
produce. But it won't be able to produce a reasonable output if it
looks at the problem through a narrow aperture.
(Maybe this should really be in IRTF?)
Ietf mailing list