Ravi at Cisco implemented the mechanism that Curtis discussed at the last NANOG. The number of flaps is now 8 flaps/second. The mechanism appears to work well in limiting the rate and the peak.
Basically, something that flaps infrequently is not surpressed for long. Something that flaps alot is surpressed a long time.
Each box will make its own decision for surpression. Sean doesn't like this since he likes the IBGP to be consistent throughout his AS. So adjustments were made to push the flap surpression out to the edges. This means that the customer boxes might see the flap TWICE (which would cause the supression to last much longer than necessary).
Within Sprintlink, there are about 150 prefixes that are frequently surpressed. There are about 12 prefixes that are flapping once a minute. There are some prefixes that gently flap over the course of a day. The mechanism does not cause this be recognized. Most of the prefixes that have these problems are /24. It also supports the notion that aggregated routes don't flap much. It is strongly recommended that /24s be statically routed at the routers' access to the core.
The code is pretty stable now and will help somewhat with the big flapping cases.
Curtis says that the draft does permit to have a second set of parameters that would permit using the dampening mechanism to capture these gently flapping routes.
Eric asked if the code dampening outbound....Sean is not sure.
Steve wants to know if Sprint can make the current list of surpressed routes is available. Sean thinks that is reasonable, but doesn't know how that information might be made available.
Tony Bates says that dampening based on ASpath is probably a bad thing and needs more research before using it.
Sean says that the default values are being use by Sprint Link.