[sidr] RIB Size Estimation for BGPSEC

Try again ...
and again ...

 RRG, before looking for some architectures that would solve the world's routing
 problems, did discuss the expected capacity of DFZ routers,  given the expected
 rate of growth of technology, so as to work out how long the IETF has got to
 come up with something new.

 The consensus was that routers could not be expected to support more than 1M
 entries in the RIB which is somewhat less than the 2.7M you expect in 2025.
And
 the new architecture could make the world look very different by then, with
 outer routing headers for delivery to location, and inner routing headers for
 delivery to the identified host.

 Tom Petch

> ----- Original Message -----
> From: "Sriram, Kotikalapudi" <kotikalapudi.sriram@nist.gov>
> To: "sidr wg list" <sidr@ietf.org>
> Sent: Tuesday, May 24, 2011 2:28 PM
> Subject: [sidr] RIB Size Estimation for BGPSEC
>
>
> In response to a request during Q&A at the SIDR WG meeting, April 1, in
Prague,
> I am posting the results of modeling and estimation of the RIB size for
BGPSEC.
> (I plan to present this work at the SIDR WG meeting in Quebec.)
>
> Here is the link for the slides:
>
> http://www.antd.nist.gov/~ksriram/BGPSEC_RIB_Estimation.pdf
> { file:///D:/Articles/BGPSEC_RIB_Estimation.pdf }
>
> I will also shortly post an excel spread sheet so you can input your view
> of the BGPSEC world (signed prefix-paths you expect to see in your PE router
> or route reflector over time), and observe the corresponding RIB size
estimates.
>
> Key observations/conclusions – mostly copied from the summary/conclusions
slide
> (Slide #26):
>
> o RIB sizes have been estimated for realistic Route Reflector (RR) and
Provider
> Edge (PE) scenarios (with input from a large, Tier 1 ISP).
>
> o Different signature algorithms considered (RSA-2048, ECDSA-256).
>
> o RIB memory needs to be engineered so it is adequate for algorithm/protocol
> transition down the road.
>
> o 15% yearly growth for # unique eBGP prefixes (377K in 2011 to 2.7M in 2025).
>
> o 9.55x multiplier applied to determine # eBGP prefix paths in RR (i.e., #
eBGP
> prefix paths in RIB = 9.55 x # unique eBGP prefixes).
>
> o BGPSEC adoption curve – truncated Normal distribution (slides 7-9).
>
> o RIB memory requirements (for the RR) are 0.5GB in 2016 to 8.3 GB in 2020 to
32
> GB in 2025 (worst case assuming RSA-2048 all the way; see Slides 14, 16).
>
> o But algorithm transition to ECDSA-256 likely (should happen) before 2020 [or
> perhaps even from Day 1 of BGPSEC deployment?].
>
> o Assuming such transition, 20 GB for RIB memory would be a conservative
> estimate –  good all the way out to 2025 (see Slide 25).
>
> Comments/suggestions are welcome.
>
> Sriram
>
> K. Sriram
> http://www.antd.nist.gov/~ksriram/
> _______________________________________________
> sidr mailing list
> sidr@ietf.org
> https://www.ietf.org/mailman/listinfo/sidr
>