Sorry, I accidentally hit Send (instead of Save) halfway through in the previous email. Here is the complete email.
Jakob, Three key specific questions one would ask about a model are: Q1. Does it model the underlying physical system accurately. In this instance, would it correctly compute the RIB size, given a number of routes (i.e., total # prefix-paths) in the RIB? Q2: Is it well rooted in the historical measurement data for its projections into the future? Q3: Are additional assumptions made for projecting into the future reasonable? In this instance, are the assumptions regarding the BGPSEC take rate reasonable? Answers (IMHO): For Q1: Yes. There is no guess work involved here. I feel we've accurately factored in the update overheads of BGPSEC based on draft-ietf-sidr-bgpsec-protocol-00. For Q2: We used up to date historical measurement data from http://www.potaroo.net/ispcol/2009-03/bgp2008.html (also Geoff's 2009 predictions in that paper came out nearly accurate through the next two years into 2010 and 2011.) We also used current measurements (as of March 2011) from a large Tier 1 ISP regarding the total # routes they observed in the PE routers and in Route Reflectors. We used the observed growth rates for IPv4/IPv6 from the past several years to project into the future. (Needless to say, predicting the future is never a hard science.) We have provided the excel speadsheet so anyone can inject their world view regarding growth rates and see what happens. http://www.antd.nist.gov/~ksriram/BGPSEC_RIB.xls For Q3: This is the hardest part of the modeling -- very tricky and no one can speak to it with certainty. We did overlay a BGPSEC take rate model (slides 5 and 6) http://www.ietf.org/proceedings/81/slides/sidr-2.pdf This was broadly based on historical view of take rates of new technologies in general. (It is usually a Normal distribution curve.) This was felt to be OK within the BGPSEC design team discussions. Once again, anyone curious can play with the speadsheet to input their assumptions/knowledge regarding their projected take rates for BGPSEC. So at the minimum, if you just wish to find out the RIB size corresponding to a given number of routes (i.e., total # prefix-paths), the model would give you that accurately. You are further at liberty to input into the speadsheet any assumptions regarding the growth rates and take-rates per your view of the world. I hope this helps some. Sriram >Do you trust this? >http://www.ietf.org/proceedings/81/slides/sidr-2.pdf >-- >Jakob Heitz. _______________________________________________ sidr mailing list [email protected] https://www.ietf.org/mailman/listinfo/sidr
