|> Adding more prefixes will hasten the inevitable. | |If it isn't justified quantitatively then it's just FUD.
Please see my RAWS talk, oh so long ago (http://www.iab.org/about/workshops/routingandaddressing/Router_Scalability. pdf). I showed (quantitatively) that the growth in the routing table simply exceeds the speed improvement of DRAM. The implication is that unless we increase cost structures, convergence will suffer. |Do the folks on this group clearly understand the harm that the |Regional Internet Registries do by denying number resources to |multihomed registrants based on our assurance that it's a necessary |evil in order to keep BGP stable? AFAIK, the RIRs are, for v4, requiring sane and reasonable justification for PI prefixes. For v6, they're effectively giving everyone PI. Can you be more specific about the policies that you're objecting to. |If we still can't reliably quantify a BGP scaling limit then our |request that the RIRs suppress BGP growth by suppressing resource |assignment is simply unconscionable. Do we need to reliably quantify where the Earth will reach the tipping point into a irreversible greenhouse effect before global warming is relevant? |> It is |apparent that unless |> there is some significant progress somehow, the cost and/or |complexity of |> running the current architecture is going to start to climb. | |Climb to what? That depends entirely on the continuing prefix growth rate and the level of additional sophistication and replication necessary to support the rate. |[Gedanken experiment with PC's for BGP elided...] | |Can we not compute the same for a $2k PC 5 and 10 years ago? If we |can, then we can with scientific accuracy project where that limit |will be for the foreseeable future. You're asking for an absolute hard limit in a situation where there are many variables that you have not yet quantified. How many peers does the BGP speaker have? What is the number of paths per route? How long is the average AS path? And does this simulation still have any relevance? And why are we limited to a $2K PC? Why can't we upgrade to server architectures? Because of this complexity, it's far simpler to point out that the underlying hardware technology (DRAM, for the case of the RIB), simply cannot keep up with our needs, even given known trends in hardware scalability. |At worst, a high-speed router consists of parallel units of lower |speed routers, a linear increase in cost tied to the data rate but not |the table size. In a modern router, the FIB is distributed to each line card, and since the FIB cost is proportional to the table size, the increase in cost is tied to both the data and size. This is the fundamental "rate * state" quandry. |So if the above can be done we will have, with |scientific accuracy, deduced a function which yields an upper bound |for BGP scalability. No, you've only driven towards a simulation that shows what constant cost might yield for a single data point on a multi-variate surface. Tony _______________________________________________ rrg mailing list [email protected] https://www.irtf.org/mailman/listinfo/rrg
