[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Preserving established communications (was RE: about draft-nordmark-multi6-noid-00)

On 28 okt 2003, at 17:00, marcelo bagnulo wrote:

I don't think it is that black and white. The routing system
could propagate bad news faster than good news

I am not sure if this is true...
I am no routing expert, but i think that withdrawing a route is inherently
slower than advertising a new route, at least in path vector algorithms.

There was a time that interdomain routing was under threat from instability caused by frequent up-down and down-up transitions. The solution was to implement "flap dampening": if you flap (down-up transition) too many times within a certain timespan, your route is ignored for some time, saving some expensive BGP and route table operations on the router in question, but more importantly in routers further upstream.

Note that not propagating good news, so that a route seems to be down when in fact it is up, isn't too harmful as there are presumably other paths available so the destination is still reachable. Failing to propagate bad news on the other hand would be catastrophic, as packets continue to flow to a place where they can't be delivered. Avoiding this situation is the number one concern when running BGP.

The problem is that to detect that there is no route to a certain, all the
alternative routes with increasing AS path are tried, which is inherently
slow.

Only if there are no alternatives. If there are, the path for those is probably not that long so they show up pretty quickly. Only when the last/only path goes down you'll see the following problem:

This would be similar to the count to infinity problem, but with
different AS path.

But we usually don't care since when our last/only path goes down we're unreachable anyway.

This implies that the time to converge is proportional to
the number of ASes in the network.

No, the number of possible paths (= interconnection exists) where the ASes to the left provide transit for our viewpoint and the ASes to the right provide transit for the source AS. This number increases exponentially with the number of transit links ASes have. So more multihoming actually hurts here. :-)