|
North American Network Operators Group Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical Re: The Big Squeeze
>To back Paul up here.. >Tests we have run would indicate that this is not _generally_ the case. >I guess it depends on your point of viewing but we certainly see an >inordinate amount of oscillation in the longer prefix ranges. >(It also depends on the parameters you associate with certain prefixes of >course ;-) >(Sorry - no facts and figures to back that up) It also depends on how you mentally group things. I agree, there are a few prefixes that have an inordinate amount of oscillation. But what group membership would I assign them? I've noticed more recently assigned network numbers seem to flap more than older network numbers. Perhaps more importantly, networks in more recently assigned AS numbers seem to flap more. Again, this is just raw observations, no real numbers to back up my feelings. Don't let the way "show ip bgp flap" happens to group networks by prefix control your mental schema. I would much rather see the decision based on how any individual route behaves rather than the 'neighborhood' the network happens to be located in. We might be willing to give a network that has been stable for a long time a few extra flaps. While brand new networks that flap would be better off being 'covered' by a shorter, more stable prefix. Is the goal to encourage stable routes? or encourage short prefixes? >For instance, take this snapshot and note the /24 from within the 153.96 >network. The flap column is the interesting one compared with other class Bs. >This is not representative but gives a flavour of what we have been seeing. > >They are all from the same originator and it would not take a great amount >of effort to apply a null0 tie down ! Or even aggregate them. But you bring up an important point. Often all the routes from the same originator flap together, irregardless of their prefix length. -- Sean Donelan, Data Research Associates, Inc, St. Louis, MO Affiliation given for identification not representation - - - - - - - - - - - - - - - - -
|