I would suggest that the largest percentage of flapping prefixes in the global routing system belong to prefixes longer than /19.
Hence the convention to damp differently for different lengths. See one of the foils in http://www.psg.com/~randy/970210.nanog/, which suggests that we over here start following the European lead on this.
Is the route computation of a /8 prefix flapping once a second any different than a /24 flapping once a second? If /8's are "naturally" more stable, then why allow them to flap more before dampening them? When dampening was first being rolled out I remember one of the early networks that got hit was PSI's net 38/8. Treating flapping prefixes differently based on length has more to do with how many people scream when prefixes covering a large amount of address space get dampened than the impact of the route flap of an individual prefix on the router. Although most folks have permanently filtered it, isn't 1/8 still the flappiest prefix of all. -- Sean Donelan, Data Research Associates, Inc, St. Louis, MO Affiliation given for identification not representation
When dampening was first being rolled out I remember one of the early networks that got hit was PSI's net 38/8. Treating flapping prefixes differently based on length has more to do with how many people scream when prefixes covering a large amount of address space get dampened than the impact of the route flap of an individual prefix on the router.
Also, it is thought that longer prefixes tend to flap more than shorter. randy
Also, it is thought that longer prefixes tend to flap more than shorter.
That's not because of the prefix length per se; it's because shorter prefixes tend to be associated with a greater number of reachable destinations per prefix, and that tends to imply better infrastructure and more opportunities for aggregation and hold-ups. --apb (Alan Barrett)
When dampening was first being rolled out I remember one of the early networks that got hit was PSI's net 38/8. Treating flapping prefixes differently based on length has more to do with how many people scream when prefixes covering a large amount of address space get dampened than the impact of the route flap of an individual prefix on the router.
Also, it is thought that longer prefixes tend to flap more than shorter.
randy
Sean has a good point here. A flap of a /8 is the same as a flap of a /24 from a computational point of view. There is clearly some social engineering going on here. If you want your long prefix to be golbally visable and you allow it to flap, then you will be subject to dampening. On the other hand if you renumber into a larger aggregate, then you are protected from dampening (to a greater degree). Kind of a 'carrot and stick' approch. :-) Erik
On Mar 3, 1997, Erik Sherk wrote:
Sean has a good point here. A flap of a /8 is the same as a flap of a /24 from a computational point of view. There is clearly some social engineering going on here. If you want your long prefix to be golbally visable and you allow it to flap, then you will be subject to dampening. On the other hand if you renumber into a larger aggregate, then you are protected from dampening (to a greater degree). Kind of a 'carrot and stick' approch. :-)
Computational power required for a route flap is not the issue here. Many people have stated that, statistically longer prefixes flap more. Unfortunately, they have then said that because of this shorter prefixes should have looser dampening parameters put on them, when what they really meant was that the longer prefixes should have more strict dampening parameters put on them. Yes it is exactly the same thing, but it is an important semantic distinction. If a group of prefixes categorized by a its length tends to flap more than the average, then said group should have more strict dampening parameters placed on it. Alec -- +------------------------------------+--------------------------------------+ |Alec Peterson - ahp@hilander.com | Erols Internet Services, INC. | |Network Engineer | Springfield, VA. | +------------------------------------+--------------------------------------+
Computational power required for a route flap is not the issue here.
Many people have stated that, statistically longer prefixes flap more. Unfortunately, they have then said that because of this shorter prefixes should have looser dampening parameters put on them, when what they really meant was that the longer prefixes should have more strict dampening parameters put on them. Yes it is exactly the same thing, but it is an important semantic distinction. If a group of prefixes categorized by a its length tends to flap more than the average, then said group should have more strict dampening parameters placed on it.
Alec
You're right - what you propose makes some sense. The reason people have proposed and are damnening on longer prefixes is: 1) To encourage people to renumber into larger (P and/or PI) space, and 2) To lessen the percentage of the net which will be temporarily unreachable by the aggressive dampener. Avi
On Sun, 2 Mar 1997, Sean Donelan wrote:
Is the route computation of a /8 prefix flapping once a second any different than a /24 flapping once a second? If /8's are "naturally" more stable, then why allow them to flap more before dampening them?
It's just another incentive to renumber into larger aggregate blocks. A provider can say to a customer: "If your /24 flaps you're going to be unreachable from some parts of the net for longer than you would if you renumber into our block." pbd
participants (7)
-
Alan Barrett
-
Alec H. Peterson
-
Avi Freedman
-
Bradley Dunn
-
Erik Sherk
-
randy@psg.com
-
Sean Donelan