North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: Why doesn't BGP...

  • From: Vadim Antonov
  • Date: Sat Nov 09 16:22:23 1996

Avi Freedman <[email protected]> wrote:

>Anyway, sorry to be snide - any suggestions you have about algorithms
>for exchanging routing data between (or inside of) providers that also
>takes into account link size and perhaps hop-count, as opposed to
>just AS-PATH, are welcome at the next IETF meeting.

The problem with physical metrics is that they're pretty much useless
in the real world, where network administrators often want to have
traffic to go along suboptimal ways to balance the load.

Obviously, any "automagic" load-sensitive system must have a global
knowledge, and long-term statistics on traffic patterns to be able
to simulate the behaviour of "gut feeling" of the engineers.

Regarding BGP metrics -- there's a problem with a single "universal"
metric in an exterior routing protocol, namely the absense of a single
authority to assign those metrics.  Different networks may have very
much confiliciting interests in how to assign BGP metrics.

Some time ago i wrote a draft (which didn't get any support and quietly
expired) regarding BGP metrics -- the idea was to have a vector of
metrics (one per AS hop) instead of a single scalar metric; and allow
network operators to calculate weighted sums to obtain the locally
significant scalar metrics.  I.e. a network operator may decide for
himself how to interpret BGP metrics coming from different AS-es
(or by different paths).

The draft also describes a mechanism for scaling IGP metrics to generate
BGP metrics (thus replacing MEDs with a generic mechanism).  It also
allows "fudging" the weighted sum with an exit-specific constant, thus
superceding local prefs.

In other words, the proposed metric system is generic enough to make
MEDs, LOCAL_PREFs, and many cases of use of communities, a mere
particular cases; so they can be obsoleted.

The default metric for an AS-hop is 1.0, and the default weight vector
is all-ones, so the default behaviour is identical to that of the current
BGP.

If there's any interest i can dig the draft out and place it on my www
server.

--vadim
- - - - - - - - - - - - - - - - -