Re: Active Route - OT

[Jeffrey Haas <[e-mail]>]

On Fri, Oct 04, 2002 at 10:54:12AM +0200, [e-mail] wrote:
> Well, I tend to be picky on the attributes. Why do we forward them, or have long
> discussions about then (as the ones I see ongoing for e.g. the AS-PATH) if in
> the end the attribute is not trustworthy?

That was what ATOMIC_AGGREGATE was intended to signal.

One of the problems with creating aggregates without full path information
is that you can't tell if there are contributors that you might loop to.
You essentially must count on the more specifics being present
or being far enough away from things to not be affected by a potential
looping situation.

Thankfully, this seems to often be the case.

Dennis points out that people largely form aggregates along their
assigned address boundaries anyway.  So long as proxy aggregation
doesn't happen all over the place, we're fine with incomplete path
information.

> What is e.g. the value of a routing policy saying that "I do not want to use
> routes that to pass via this particular AS".

The fact that it lets you shape your packet flows.  Even a coarse
control is better than none.  And when none is available, people
will sometimes internally de-aggregate in order to get that control.

> Unless  of course, my whole appreciation of the atributes is wrong. But in this
> case it would be good to indicate what can and cannot be expected from an
> attribute.

Honestly, its best to *not* expect anything beyond guaranteeing loop
free routing.  The fact that you can *try* is an operational decision.
It might even work.

> So we would probaby end up with an indicator in the route saying " attributes
> not guaranteed", but then again i think i prefer the restriction in the RFC.

We could also say that everyone must propagate every route that enters
the routing system.  I rather suspect network operators and other
implementators would tell us where to put that suggestion.

-- 
Jeff Haas 
NextHop Technologies