From: Iljitsch van Beijnum (no email)
Date: Wed Jan 02 2008 - 18:09:41 EST
On 2 jan 2008, at 22:34, Joe Abley wrote:
> The community who would like the knob not to be "deaggregate" are
> the same ones that are doing the deaggregation, which I think is as
> it should be from a macro level
More precise: the two sets of people are part of the same community.
I'm not sure if there's much overlap between the really bad
deaggregators and those who are strongly pro-knob, though.
> As to "there must be better knobs" I think it may be a little late
> for that; by design (or as a consequence of it) the set of IPv6
> knobs is the same as the set of IPv4 knobs.
The trouble is that BGP doesn't have a meaningful inter-AS metric.
(Although there is something that is called that.) If I want to
increase my path length by 10% through a certain neighboring AS, I
don't get to do that. I only get to double or triple it. (Unless I was
doing very heavy prepending to begin with.)
This is easy to fix by adding a new metric to BGP that is increased by
10 or 100 or 1000 at each hop by default, but which can also be
increased by a larger or smaller amount as desired. In essence, this
would make the AS path a lot more granular. Obviously this only works
if a fairly large set of ASes implements this.
However, word on the street is that in order to get a new BGP
attribute defined in the IETF idr wg, you need assurances up front
that people are actually going to implement and use that new attribute.
|
|
|