draft-hardie-bounded-longest-match-01.txt

[Jeffrey Haas]

I was reading through draft-hardie-bounded-longest-match-01.txt and
was struck by something.

The base BGP document declares in strong terms:

:   The function that calculates the degree of preference for a given
:   route shall not use as its inputs any of the following: the existence
:   of other routes, the non-existence of other routes, or the path
:   attributes of other routes.

Ted's draft specifically changes this.

Does anyone know the original reasons for this section in the BGP document?

As Ted notes in his draft, route churn/explosion is probably the biggest 
reason.  Removing a covering aggregate may lead to several "new"
prefixes to show up in a otherwise "stable" routing system with the
relevant convergence effects on the entire Internet.

I would also note that "How Might this work" covers many of the same
checks needed to properly implement ATOMIC_AGGREGATE.  I don't
believe ATOMIC_AGGREGATE is implemented by many vendors.  This is
likely due to the extra overhead involved.

Comments?

-- 
Jeff Haas 
NextHop Technologies