Short version: I think constraints 1 to 7 are absolute. If the
best solution can't fully meet constraints 8, 9 and
10, then maybe it could still solve the routing
scaling problem to a worthwhile degree.
Hi Noel,
Thanks very much for this:
>> http://www.firstpr.com.au/ip/ivip/RRG-2009/constraints/
>
> I only have time for a quick skim of this (_way_ behind :-), but the list of
> points generally looked pretty good to me.
>
> My only concern was that, in toto, the entire list might be the equivalent of
> asking for a car that goes 400 KPH and gets 500 MPG (or whatever the
> equivalent is in kilometres per liter, sorry everyone in the rest of the
> world :-), i.e. we can't have all of them at the same time. So it might be
> that we'd have to prioritize some as being more important than others.
I think constraints 1 to 7 are absolute. Here are my thoughts on 8,
9 and 10.
If the solution did not meet constraint 8, this means that could
still be highly attractive for many smaller networks which haven't
yet started using the unscalable approach to multihoming etc. (PI
space advertised in the DFZ), but would be either less attractive or
not attractive at all for those networks which already have their own
PI space.
For instance, maybe the best proposal wouldn't be adopted widely by
networks with PI space, so the current number of DFZ routes would not
decline, but at least it would not grow as fast as without the
solution - and millions of smaller networks will get multihoming,
portability and TE on a scalable basis.
Constraint 9 - not compromising security, robustness or performance -
is a sliding-scale affair. Ideally there would be no compromise.
If the best we can do involves some compromise, then perhaps we can
still get wide enough adoption to solve the problem and provide
multihoming etc. for the large number of smaller networks which can't
get it now.
There is an argument that smaller networks, which couldn't afford PI
space, even it was available (and it is for IPv6) wouldn't be
contributing to the scaling problem anyway, and don't have any
alternative for multihoming etc. than to adopt the scalable solution.
In that case, *maybe* they would accept some degradation in
security, robustness or performance.
If and when IPv6 or some other addressing scheme becomes useful for
end-user networks as an alternative to IPv4, then I think it can be
assumed that the high costs of PI space, and of advertising it, which
are the pattern with IPv4 won't be replicated with IPv6 or whatever.
So in a broader architectural sense, beyond IPv4 (I am yet to be
convinced we ever will move beyond IPv4, except for cellphones), we
can't rely on the unscalable BGP approach to multihoming etc. being
"expensive".
I think that any substantial perceived degradation (whether real or
not) will make it impossible to gain wide enough adoption as long as
the unscalable alternative is available to each end-user network.
If the scalable solution offers second-rate performance compared to
conventional PA or PI space, or is perceived as being unsuitable for
large networks, then many small networks will avoid it - especially
smaller networks which plan to be larger in the future.
Constraint 10 is a sliding scale one too. Widespread voluntary
adoption could best be achieved if there was no need for any internal
router upgrades or replacements. However, for small networks, there
may be only one router and its cost may be so small as to present
little impediment to widespread adoption. For larger networks,
having to replace many or all internal routers - presumably expensive
ones - would surely be a barrier to voluntary adoption which would
make it much more difficult to wean these networks from their current
unscalable multihoming and portability arrangements.
- Robin
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
