Dampening considered harmful? (Was: Re: verizon.net and other email grief)

Well, a particular router doesn't get to set its dampening according to its 'view' today, and that view is going to vary depending on prefix.

I would like to argue that how we define flapping today is simply a broken concept. We count up/down/path change transitions, but such transitions can exist due to connectivity or implementation differences and may have nothing to do with not complying with ettiquette.

If people were damped for a single up/down, then yes, that's a problem. Dampening someone because they have a single path change is not the idea. That would sort of nullify the spirit of being multihomed with BGP, or even using BGP in any production or transit network for that matter.

IF there's a connection problem, or implementation difference that makes a lot of up/down, then dampening could occur close to the "problem" but it will be contained close, and won't spread to the rest of the internet.

I submit that a better way of measuring flap is to look at the period over which a particular prefix is flapping. Anti-social behavior is many changes occurring over long periods of time. However, many changes over a short period may well not be a problem.

Regards,
Tony

So configure your route dampening that way. No one's twisting your arm to use the defaults. Want many flaps in the short term to not be damped? Simply increase your suppress limit. Way way up. Increase or decrease your half-life as you see fit. Just don't expect the rest of the world to do the same. But do expect to see others damp the flapping that you leak to their network.

<gut feeling> If no one on the net ran dampening, routers wouldn't get anything done except processing constant churn. </gut feeling> I'm grateful that routers out there in the world are doing route-damp so my router's CPU doesn't have to deal with needless route-churn, even if it does impact my connectivity (to poorly connected) end sites somewhat. The stability is worth it.

Another item to consider, is this is a sentence from RFC2439 " By damping their own routing information, providers can reduce their
    own need to make requests of other providers to clear damping state
    after correcting a problem."

Using that logic, I'm also grateful for my transit providers dampening the advertisements to their peers, of my flapping transit links to them. That way, when they flap, (hey, it happens to everyone eventually) it doesn't get me damped by the rest of the internet, and my connections to different transit providers can do their job of carrying the traffic.

Dampening, in it's current state, really is a good idea.

-Jerry

Today's AS hierarchy is quite flat, which severely limits the usefulness of dampening. If the link between ASes A and B flaps, then B doesn't get to dampen these flaps. C, connected to D, does, but if C is a small network that doesn't help much as flap dampening brings its own overhead. In a two or three router network there probably isn't any advantage in dampening. Only when you get to protect a larger number of routers from the update, it helps. Now of course D, connected to C, will be isolated from the instability. But in today's internet, there often isn't a D. According to the weekly routing table report the current average AS path length is 4.5. Subtract at least .5 for prepending, and there must be a significant number of 3 or even 2 AS hop paths to get the average at 4.