Elliott Karpilovsky wrote:

Hello everyone. My name is Elliott Karpilovsky, a student at Princeton University. In collaboration with Alex Gerber (AT&T Research), Dan Pei (AT&T Research), Jennifer Rexford (Princeton University), and Aman Shaikh (AT&T Research), we studied the extent of IPv6 deployment at both global and local levels. Our conclusions can be summarized by the following three points:

1.) IPv6 deployment is not seen as a pressing issue. 2.) We saw a lack of meaningful IPv6 traffic (mostly DNS/Domain and ICMP messages), possibly indicating that IPv6 networks are still experimental. 3.) Studying Teredo traffic suggested that it may be used for NAT busting by P2P networks.

Our paper (submitted and presented at PAM 2009) can be found at http://www.cs.princeton.edu/~elliottk/ipv6study.html . If you have comments or feedback with respect to these results, please feel free to express them.

Nasty comment time... "To analyze native IPv6 traffic, we use Netflow records collected from an IPv6 Internet gateway router in a US tier-1 ISP with 11 IPv6 BGP neighbors. These records were collected from 2008-4-1 to 2008-9-26, and are taken from the business customers. " Sorry to have to make this comment, but the IPv6 side of the Internet is quite a bit larger than "11 peers". I don't really think that AT&T can call themselves a "tier-1 ISP" on the IPv6 field (they can on IPv4), especially as there are these wonderful give-aways as using OCCAID as a transit: [..] 7 fr-par02a-re1-t-2.ipv6.aorta.net (2001:730::1:2d) 51.944 ms 51.596 ms 51.915 ms 8 uk-lon01a-re1-t-1.ipv6.aorta.net (2001:730::1:2a) 60.802 ms 61.405 ms 61.498 ms 9 ibr01-ve26.lndn01.occaid.net (2001:7f8:4::7577:1) 37.941 ms 37.797 ms 37.88 ms 10 bbr01-p1-0.nwrk01.occaid.net (2001:4830:fe:1010::2) 106.622 ms 106.538 ms 106.701 ms 11 r1.mdtnj.ipv6.att.net (2001:4830:e2:2a::2) 145.847 ms 145.762 ms 146.049 ms 12 2001:1890:61:9017::2 (2001:1890:61:9017::2) 222.045 ms 222.694 ms 223.185 ms 13 mail.ietf.org (2001:1890:1112:1::20) 221.683 ms 221.66 ms 222.839 ms Heck, I can't find a single ISP in GRH with which I can reach AT&T (where eg www.ietf.org is currently in) from Europe directly. Unfortunately, I will have to state that that thus completely makes that whole paper useless as the data is used is just that: useless. I really really really hope that AT&T finally realizes that they have to start deploying IPv6. When they have done that, re-run your "study" and then release those numbers as then they will maybe be interesting when there are actual customers on the links. Greets, Jeroen

On Mon, Apr 27, 2009 at 10:56 PM, Elliott Karpilovsky < elliottk@cs.princeton.edu> wrote:

Hello everyone. My name is Elliott Karpilovsky, a student at Princeton University. In collaboration with Alex Gerber (AT&T Research), Dan Pei (AT&T Research), Jennifer Rexford (Princeton University), and Aman Shaikh (AT&T Research), we studied the extent of IPv6 deployment at both global and local levels. Our conclusions can be summarized by the following three points:

1.) IPv6 deployment is not seen as a pressing issue.

Agreed. SPs are driven by customers. Customers, generally, still want the IPv4 net. However, at least where I am at, we have started to gain more and more demand for IPv6 services (in this case, its specific to Private IP services). The relief is hampered by the ability to provide the service quality demanded by our customers. As an SP, if you can't provide the quality and technology together, you will push back to stave it off until you are able to provide both (either way is less than optimal, but one way results in losing whole accounts and the other is just a minor setback)

2.) We saw a lack of meaningful IPv6 traffic (mostly DNS/Domain and ICMP messages), possibly indicating that IPv6 networks are still experimental.

There is not a widespread adoption yet. Until SPs are deploying gear that is adept to handling this traffic and able to guarantee the service quality, there will not be a significant load on the IPv6 infrastructure. On a separate rant, since we have to NAT/PAT on IPv4 already, who really cares if we NAT/PAT between IPv4 and IPv6? Interop as a transition tool would certainly hasten the deployment of IPv6. With major SW vendors now providing full support for the IPv6 suite, SPs that provide interop with IPv4 can start the migration sooner rather than later.

3.) Studying Teredo traffic suggested that it may be used for NAT busting by P2P networks.

Kudos to the p2p developers/users who have gone this route. What an intuitive way to handle matters. On Tue, Apr 28, 2009 at 4:06 AM, Jeroen Massar <jeroen@unfix.org> wrote: Unfortunately, I will have to state that that thus completely makes that

whole paper useless as the data is used is just that: useless.

I really really really hope that AT&T finally realizes that they have to start deploying IPv6.

When they have done that, re-run your "study" and then release those numbers as then they will maybe be interesting when there are actual customers on the links.

Greets, Jeroen

From our perspective as net engineers, this is how we are going to view this. The information in the document gives *some* good information, but Jeroen is right... the data coming off of AT&T nodes doesn't give any credence to the report. The report *does* tell us that there is still an active effort to avoid IPv6. The rationale used to derrive the conclusions in the report is lacking at best, harmful to adoption at worst. I feel that the grossly incomplete data will be percieved as a lot of FUD coming off this report, but I'm unsure who it would benefit to maintain such stances (except a current Tier-1 IPv4 provider who doesn't have the same status in the IPv6 Internet).