Re: Dampening considered harmful?

[Jerry Pasker <info () n-connect net>]

> Back in mid-December someone typed:
>
> >  > One reason to be careful with dampening is that flaps can be
> >  > multiplied. (Connect to routeviews and see the different flap counts
> >  > under different peers for the same flap at your end to observe this.)
>
> How about in this scenario:
>
>   asA gets transit from asT
>   asA gets backup transit (ASpath padding) from asB
>     asB gets transit from asT
>     asB gets transit from asJ
>       asJ gets transit from asT
>         asT peers with whole world(*)
>
> Now, as asA flaps to asT, we see "bad things" happen to their routes,
> namely an unreasonable amount of flap at even nearest neighbors to asT.
>
> Can this flap magnification be explained by the hierarchy I describe
> above?  That is, asT treats all of these ASpaths as customer routes:
>
>   asA
>   asB_asA
>   asJ_asB_asA
>
> and so we might expect to see multiple flaps as different "best"
> routes come into view inside the geographically diverse asT...  right?
>
> Thanks,
> -mark
>
> (*)you know what I mean.

I post this response to the list also seeking clarification/further education.

The PATHS that flap at each router get damped, right?  So if a path 
to an AS doesn't flap (best path or not), then it doesn't get damped. 
So if a best path gets damped, suddenly it's no longer "best" and 
another non flapping path to that AS becomes the new best.

I had to draw this hypothetical network out on a white board, and 
then it became a little easier to "see".

If the path from A to T flaps into oblivion, the path from A to B 
starts carrying traffic from A<-->B and B<---->J> and  J<---->A (via 
B).  The A to T link can flap all day long, and B and J will still 
have undamped connectivity to A.

T will send flaps to J and B and J and B should damp those paths that 
are flapping, and continue to send traffic to A via the links that 
aren't flapping.  The PATHS that flap at each router get damped, 
right?  If that's the case, then eventually, everyone will damp all 
the "A" flap from T, and start to prefer any paths  to A from J and B 
(because they aren't flapping, and therefore are better, more stable, 
and more preferred paths)   If T is the only transit provider that's 
peered with everyone* then A might have a problem.   Perhaps A should 
chose their transit providers a little more carefully.  If A, B, and 
J are all tier 2 ISPs peered with each other that get transit 
ultimately from T, then none of them are really truly multi homed. If 
J was also 'nearly peered with everyone' then eventually, A traffic 
would all start to arrive via J through the connection to B.

If all the above is correct, then dampening just made the network 
much more stable in the above case, rather than traffic constantly 
routing to and them re routing around the flapping link from A to T.

Stability good. Packet loss and latency, bad.  Stability = damp.

-Jerry