1. Its The End Of The World As We Know It (aka The New Internet Architecture) David Meyer (Speaking for myself) IEEE 22 nd Annual Computer Communications Workshop Steamboat Springs, CO 23 October 2008

2. ITEOTWAWKIIEEE CCW08Slide 2 Before We Dive Into All Of This… Notwithstanding reports to the contrary, the sky hasnt fallen (yet)

3. ITEOTWAWKIIEEE CCW08Slide 3 Agenda Whats the Problem? What Is the New Architecture? –And what world is ending? What My (very cloudy) Crystal Ball Tells Me So Whats Next? Q&A

4. ITEOTWAWKIIEEE CCW08Slide 4 Whats the Problem? Data Plane under attack due to panic based on IPv4 run out –O(30) /8s left in the IANA Free Pool –Dual-stack transition to IPv6 abandoned –All varieties of NAT (and beyond) being proposed Control Plane under duress (crumbling?) due to various operational practices and economic concerns –And we havent even seen widespread adoption of IPv6 –That combined with the RIRs IPv6 PI-for-all allocation strategies means more O(rate*state) in store for the control plane –And deaggregation is on the rise (for various reasons) –Ill just note that the IPv6 designers never dealt with control plane issues (or at least, not successfully)

5. ITEOTWAWKIIEEE CCW08Slide 5 BTW, What Do The RIR IPv4 Allocations Really Look Like?

6. ITEOTWAWKIIEEE CCW08Slide 6 Ok, But Whats The Concern? Assertion: The lack of a reasoned approach to both the IPv4 run-out problem (data plane) and the growth of routing state (control plane) are life-threatening to the (end-to-end) Internet we all know and love I want to focus on the data plane (because thats the panic de jour), but lets overview the control plane issue for a minute….

7. ITEOTWAWKIIEEE CCW08Slide 7 Internet Control Plane Whats the Issue?

8. ITEOTWAWKIIEEE CCW08Slide 8 Is Locator/ID Split the Solution? Changing the semantics of the IP address LocatorID 2001:0102:0304:0506:1111:2222:3333:4444IPv6: 209.131.36.158IPv4: Locator.10.0.0.1 ID ID & Location If PI, get new locator If PA, get new ID

9. ITEOTWAWKIIEEE CCW08Slide 9 Scaling the Control Plane Lots of solutions based on the Loc/ID split idea –See Noels page for some of the definitive papers on this idea –Basically, you have one (blunt) instrument: topological aggregation LISP, 8+8/GSE, Six/One Router, IvIP,… None has seen serious implementation other than LISP, and none has seen serious production deployment See http://www.lisp4.net for some information on the LISP network So lets get on to the data plane

10. ITEOTWAWKIIEEE CCW08Slide 10 Scaling Internet Data Plane: What Was The Plan? IPv6 –But, how to get there from here? Well, we had Dual Stack, and we had… Dual Stack, and we had… Dual Stack Dual stack turns out to be an inherently flawed approach –While I can signal that the correspondent host is IPv6 capable with a AAAA record –This tells me nothing about the capability of the data path –Consider Vistas behavior

11. ITEOTWAWKIIEEE CCW08Slide 11 Dual-Stack IPv6 Uptake Model* * Graphic due to Geoff Huston

12. ITEOTWAWKIIEEE CCW08Slide 12 What Really Happened* * Graphic due to Geoff Huston

13. ITEOTWAWKIIEEE CCW08Slide 13 So What Is The New Internet Data Plane Dual stack as a transition mechanism has been effectively abandoned –draft-arkko-townsley-coexistence-00.txt IPv6 uptake has been, well, disappointing So what do we see emerging? –Carrier Grade NAT (really, really big double/triple NAT) –A+P (Steal some bits from the port) –Dual-stack Lite (use NAT+tunneling) –IVI (Embedded Address Protocol Translation) –IPv6 edge with IPv4 core (tunneling) Does this work? ipv6.google.com –…–… draft-wing-nat-pt-replacement-comparison-02.txt

14. ITEOTWAWKIIEEE CCW08Slide 14 Crystal Balls and the Like Carrier Grade NAT will be deployed –After all, this is red-meat for vendors Dual-Stack Lite will be standardized –Recently added to the softwire WG charter as a work item A+P will be picked up by those who dislike CGN Potential for IPv6 to be confined to the edge –Which could be fatal for IPv6 –Can IPv6 survive as a purely edge technology? –And does that solve any problem? So what are we left with?

15. ITEOTWAWKIIEEE CCW08Slide 15 Crystal Balls and the Like Ok, Carrier Grade NAT (et al) will be deployed –Get used to it But what does it mean? –End-to-end fragile, if at all –Enormous CAPEX spins and out of control OPEX for SPs This will induce an structure on the SP industry –In the same way that super-linear growth of control plane state does And where can we look for solutions –Think the IETF or the RRG is the right place?

16. ITEOTWAWKIIEEE CCW08Slide 16 If a Picture is Worth 2^10 Words…

17. ITEOTWAWKIIEEE CCW08Slide 17 So Whats Next? For SPs, the Internet is about to become a lot more expensive to deploy and operate For Users, the Internet is about to become a lot more expensive and a lot less reliable –And a lot more balkanized Whats needed? –Serious research into what we can deploy effectively in the near-to-medium term to combat these effects (in a scalable manner) –Serious research into what kind of Internet-scale data and control planes can be designed and importantly, deployed –Coordinated/cooperative effort across a wide variety of disciplines

18. ITEOTWAWKIIEEE CCW08Slide 18 Q&A Thanks!