T-Mobile USA IPv6 Deployment IPv6-only Mobile Perspective

1. T-Mobile USA IPv6 DeploymentIPv6-only Mobile Perspective Cameron.Byrne@T-Mobile.com May 2012

2. News Flash! • T-Mobile IPv6 Deployment COMPLETE. • Android Phones are starting to support IPv6 • Nexus S • Galaxy Nexus • 464XLAT code submitted • https://android-review.googlesource.com/#/c/34490/ • ROM supporting 464XLAT and dual-stack WLAN tether http://dan.drown.org/android/clat/

3. Objectives • Listen for good ideas • More technology cooperation's and aligned strategies • Share perspective • IPv6 is needed for the Internet to grow • Dual-stack is not the answer (always…) • IPv6-only can work • User experience is key

4. Agenda • IPv6 is deployment is critical and possible • NAT is bad • Quick Review on Mobile Architecture • T-Mobile USA’s Path • Beta Deployment Results • IPv6-Only Service Enablement with 464XLAT

5. Define desired result, and work backwards • Problem: Global IPv4 exhaustion • Target: End to end IPv6 End to end IPv6 End to end IPv6 + NAT64/DNS64 for ~50% of flows (Possible today) Squat-space IPv4 + NAT44 (Yesterday) End to end IPv6 + NAT64/DNS64 for long tail

6. A few big fish make IPv6 possible Bing, Google, Yahoo, Facebook… Grandma’s blog

7. We are out of IPv4 and IPv6 is not deployed 3 February 2011 APNIC Out 2002 FreeBSD, Solaris, Linux, Windows all have IPv6 support  31 January 2011 IANA Out

8. http://www.potaroo.net/tools/ipv4/plotvar.png

9. Everyone agree IPv4 is a dead-end for “strategy”? • Mobile • Grid (m2m) • Cloud FAST GROWING EDGES THAT CAN ONLY GROW ON IPV6

10. Amazon is big, and how many servers in Google? And why did Microsoft BUY IPv4 addresses?

11. Machine-to-Machine Traffic to Increase 40-Fold Between 2010 and 2015

12. Nearly 6 Billion Mobile Subscribers

13. Over 1 Billion Mobile Broadband Subscribers • 45% annual growth over the last 4 years • Mobile broadband subscriptions outnumber fixed by 2:1

14. Mobile Traffic Growth

15. Review • Mobile • Grid • Cloud FAST GROWING EDGES THAT CAN ONLY GROW ON IPV6 We all have to engineer for IPv4-only, IPv6-only, and Dual-stack users and services

16. Total of 4.3 Billion IPv4 Addresses?

17. So, what should we do about this problem of already having 4x more connected devices than IPv4 addresses?

18. Dual-stack is not bad… but ... An imperfect Analogy Hybrid 44/44 $24k (52% premium over base) Tesla all-electric Model S No gas, $50k, 0-60mph in 4.5 seconds “Compared to a vehicle like the BMW 535i, Model S will save its owners approximately $8,000 over five years in fuel costs alone.” Standard 28/39 $15.8k

19. 20 Connectionshttp://www.nttv6.jp/~miyakawa/IETF72/IETF-IAB-TECH-PLENARY-NTT-miyakawa-extended.pdf

20. 15 Connections Buy Now

21. Anyhow, lets talk about mobile • IPv6 in mobile network is easy, its all just tunnels • Mobile has NAT today, NAT is bad • Mobile has squat-space and overlapping IPv4 – this is a big problem for IMS / SIP / WebRTC / P2P • IPv6-only in mobile is feasible with NAT64/DNS64 and has full IPv4 service parity when using 464XLAT • After world IPv6 Launch, over 50% of T-Mobile’s traffic for IPv6 users will be IPv6 end-to-end… NO NAT

22. From a network perspective, what does a smartphone want? • Constant connection to 30+ clouds for updates • push notifications, advertisement streaming, real time communications

23. Quick Architecture Briefing IP Anchor Gn Aggregation + Mobility Gb Cell Site

24. T-Mobile USA’s Path to IPv6(as a new service, incrementally per handset model deployment) • Dual-stack + NAT44 • Drives 2x the PDP and thus 2x the cost directly via contract or indirectly via utilization (bearer setup, mobility events …) • Relies on IPv4 addresses that legitimately are not available • Two different transports make troubleshooting at the user level more difficult, harder to isolate the variables • IPv6-only + NAT64/DNS64 • Cost neutral for packet core (single PDP) and drives down cost of NATs as IPv6 content goes up (AAAA) • Familiar architecture to today (1 PDP + NAT function) • Enhances current NAT with DNS64 load steering functionality, NAT no longer must be “on path” • Positively incentivizes use of IPv6 in the content network to by-pass NAT

25. Brief on how NAT64 / DNS64 workswww.viagenie.ca/publications/2010-06-03-terena-nat64.pdf

26. Impact to Network Entities

27. Findings From the Beta • Most things works fine • Web, email, … work fine. No user impact • 85% of Android apps work fine, similar general experience with Symbian market (Ovi) • Apps are developed in modern SDKs with high-level APIs • Some things don’t work • Peer to peer communication using IPv4 referrals (Skype, MSN, …) • IPv4 literals http://10.1.1.1 • IPv4 sockets APIs

28. What breaks in IPv6-only environments? goo.gl/z3j3q

29. The Face Of IPv4 Literals

30. How to make EVERYTHING work on IPv6-only?http://tools.ietf.org/html/draft-ietf-v6ops-464xlat-00

31. 464XLAT Traffic Treatment

32. Opt-in vs. Opt-out of IPv6 • IPv6-Only has limitations today for legacy applications on • advance handsets, netbooks and laptops like Skype • But, it is a good choice for basic data devices that • focus on web and email • IPv4 is not going away. IPv4-only + NAT44 will be maintain as long as needed, but we need to • start the migration to IPv6-only where • it fits. We cannot let corner cases stop • the progress of IPv6. We cannot wait • perfect.

33. References • [1] 464 IETF Draft http://tools.ietf.org/html/draft-ietf-v6ops-464xlat-00 • [2] T-Mobile USA IPv6 Beta http://goo.gl/HGmsy or https://sites.google.com/site/tmoipv6/lg-mytouch • [3] Open Source 464XLAT CLAT implementation on Android http://code.google.com/p/android-clat/ and http://dan.drown.org/android/clat/ • [4] T-Mobile USA Android Market Application Functionality http://goo.gl/z3j3q or https://docs.google.com/spreadsheet/ccc?key=0AnVbRg3DotzFdGVwZWlWeG5wXzVMcG5qczZEZloxWGc • [5] Write up on the T-Mobile USA IPv6 Service http://goo.gl/W55YQ or http ://www.androidpolice.com/2012/01/29/t-mobile-usa-testing-ipv6-on-select-devices-here-is-what-it-all-means-and-yes-no-more-nat/

34. Backup Slides

35. Things to Keep in Mind • Content, Applications, and Services will be better served via end-to-end IPv6, not NAT44 or NAT64. • In 3GPP networks, the UE always request IPv4, IPv6, or v4v6 at the start of each sessions. All 3 types can coexist which allows migration to IPv6 on a device by device basis • IPv4 Literals are a bad practice and will break our shared customers • IPv4 address embedded in HTML / XML or applications, issue explained well, but no good solution • Commonly found in video streaming content • IPv4 Literals represent major breakage in the IPv6-only network. • PLEASE USE FQDNs SO DNS64 CAN WORK, FQDNS ARE EASY AND NOBODY WANTS CONTENT TO BREAK • Customers will not tolerate broken content, they will move on to content that works. Best for the content owner to control the IPv6 experience by providing native IPv6 services, and not depend on CGN / LSN • Must be mindful of security, but security != statefulnetwork firewall • Digital divide considerations

36. IPv6 is a requirement for LTE backhaul networks, and will open opportunities for truly flat mobile networks

37. And remember how we talked about innovation and restoring the e2e principle

38. IPv6 is a Priority • All T-Mobile users, with few exceptions, have non-routable addresses and NAT44 to the Internet • Very difficult to scale and manage NAT44 capacity, AJAX and other web technologies driving hundreds of sessions per user (see NTT report …) • Large operators have to choose between BOGONs or overlapping RFC1918 space, each have pros and cons • T-Mobile USA has limited public addresses • VZW has 30x the public IP space that T-Mobile USA has, but T-Mobile is very good at LSN • NAT44 and can sustain current session growth for another ~5 year • CGN/LSN is a risk to Femto cells, UMA, and FMC in general IPv6 is needed to continue growth and avoid scaling issues on NAT44

39. The IP Address Environment • IPv4 end-point addressing is exhausted, majority of TMO mobile data users have BOGON addresses with NAT44 to keep pace with mobile data subscriber growth • Smartphones are driving a very large amount of network signaling as they create and teardown bearers • Data subscribers keep growing as well as the amount of time a subscriber is attached to the data network (always-on) • EOY 2012, projections indicate ~50% of user traffic can be served by IPv6 end-to-end -- World IPv6 Launch • Any change to handsets drives substantial cost and drives out lead time • There is no traditional business case for IPv6, have to rely on less compelling story about business continuity and exposure to upside of innovation in IPv6 combine with low cost to deploy • Leaders like Microsoft and Comcast have shown IPv6 is relevant