dean@av8.com (Dean Anderson) writes: > We are still not out of IPv4 address space. I don't know the projections > for address space depletion, but I think the IPv6 problem won't be so much > address space depletion as route space depletion. I think it will be > unwieldy to have a million or 2 million or 10 million distinct routes. > Aggregation won't happen the way things are currently organized. > - "route space depletion" seems to be an argument postulated on the idea that exterior routing scalability depends mostly on the number of prefixes carried. I tend to hold the somehow heretic view that that is not really the case. imho, exterior routing scaling depends much more on the number of distinct topologic information that is carried (i.e. distinct BGP path attributes). This much is subject to experimentation: take N distinct prefixes and advertise them in BGP w/ the exact same path attributes; compare the result w/ the same N advertised w/ different attributes. I believe that you will find that using a reasonable implementation that not only the latter consumes more memory but it also does consume a significant higher ammount of CPU resources. I would also argue that contrary to the pre-CIDR years, todays networking equipment is "shorter" on CPU than on memory. Note that topologic information (path attributes) is already the smallest ammount of information required to express the "problem at hand". Thus routing information reduction should probably focus much more on the aggregation of topology information, when possible, than reducing the number of prefixes. In current internet routing tables, the number of prefixes to distinct path attributes is usually around [5-10] interval. So it is often arguable how much effort an implementation wants to put on concentrating on the basic information (path/topology attributes). One can reasonably expect that if the ration of prefixes to attributes where to increase substantially more effort would be made to concentrate on this topologic information. Aparently this notion that path attributes may be more relevant to exterior routing scaling that number of prefixes is actually gaining popularity. See for instance http://www.caida.org/projects/routing/atoms/ Of course there is a number N of prefixes that will overload your memories, both in terms of routing and forwarding databases. But i believe it is rather simplistic to assume this is the main scaling factor w/ exterior routing. There is another point i would like to raise... today's top of the line networking equipment has the requirement to do 40Gb of traffic on a single line card, which should translate roughly to ~100 Mpps. The corolary of some of the arguments applied to the routing problem would be to say that since it is impossible to do 100Mpps using a 3-year old general purpose CPU we should all go home and declare this packet switched experiment to be utterly unscalable. The clever reader will point out that there is an economic value attached to forwarding of packets while there isn't one attached to carrying routes per se. But it is sort of impossible to determine the scale to which you can engineer a given problem without considering the ammount of resources you can throw at it... Regardless of how you skin the cat, the discussion is going to come back to the forum of economy and politics, be it in terms of how much you can scale the routing systems or in terms of shifting some of the burden to the end user networks via NAT, aggressive renumbering or any other proposal i'm familiar with. This rambling is rather long already, but i wouldn't want to end with out taking my turn at the crystal ball... Neither address or routing space exaustion seem to be realistic enought scenarios in the medium term. There are plenty of economic incentives for all involved (SPs, end users, vendors) to find solutions that allow for anyone of the planet to use the main services the internet is supposed to provide today: https and a spam-full inbox. regards, Pedro.