Hey Brian, I actually agree with you that it is difficult to predict the reasons behind deaggregation, and how much each reason has contributed to overall expansion. I also agree that even if we could somehow discover this, the past is not certain to properly predict the future. Which is why in my original message, i said:
"I wonder if it would be preferred to have a solution that allows networks to shield themselves from routing table bloat, regardless of the causes behind long-prefix announcements - a cause-agnostic solution." My last message meant to convey that IF we knew of a set of reasons that, were they eliminated, would solve our scaling problems forever, then it might help us formulate a strategy to scale the Internet. But since, as you pointed out, we cannot, we should consider more cause-agnostic solutions. Many rrg proposals are not of this flavor. I see a lot that point to TE, or PI addressing, and focus on enabling parties to receive the benefits in other ways, then claiming it solves our scalability problems. There is an underlying presupposition that if you eliminate these causes, the routing tables will scale again, and that supposition is exactly what you and I are a bit uneasy about. Dan Jen On Sun, 2009-12-06 at 09:29 +1300, Brian E Carpenter wrote: > Dan, > > On 2009-12-05 19:09, Dan Jen wrote: > > Hello, > > > >> > All of the recent proposals and rrg discussion seem to focus on > >> > re-empowering aggregation by attempting to pinpoint the reasons for > >> > deaggregation and dealing with them. A lot of reasons have been > >> brought > >> > up, such as PI usage, ingress TE, avoiding renumbering, etc. Yet, it > >> is > >> > unclear how much each 'cause' contributes to overall deaggregation. > >> > >> Does it matter? As long as the solution handles all (or most) of them, that > >> should be all that's needed. Are there significant causes of expansion that > >> aren't handled by separation of location and identity? If so, I think we > >> need > >> to hear about them ASAP. > >> > > > > Here's why it might matter. Some proposals address some, but not all of > > the alleged causes of expansion. Imagine if we knew how much each cause > > contributed to expansion, and realized that one or some of those causes > > didn't really contribute much to expansion at all. Proposals that > > didn't address these 'problems' wouldn't be subject to such a criticism > > anymore. > > > > On the flipside, if one of these causes was the found to be the main > > contributor of expansion, we certainly would make solving this problem a > > MUST. > > I'm not sure that follows. Although there is some long term smoothness > in DFZ growth vs Internet size, this conceals many non-linear effects > of the way the topology has evolved. That makes predicting the future > from the past very hard; as the recent Internet Observatory report > shows, the topology is still changing in nature (and the causes are > economic as much as they are technical, so even harder to predict). > > If you can't predict quantitatively the effect of each cause, it's > hard to prioritize which causes to tackle. > > That said, do you have any ideas on how to analyze the historical data > to clearly separate and quantify the effect of each cause of DFZ growth? > > My personal view is that we should focus on quite high level goals > that don't depend on topological or business details. Quoting myself: > "in addition to ... minimising the number of prefixes originated > by each AS, it is also highly desirable that the large majority of > ASes should continue to be origin-only systems providing no transit." > > Brian > _______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
