RIPE 71 IPv4 Report

There were three presentations of interest to me at the RIPE 71 meeting in Bucharest, Romania and some interesting information from the Regional Internet Registry (RIR) updates.

The first presentation was by Geoff Huston, the Chief Scientist at APNIC. He compared the IPv6 failure rate to the IPv4 failure rate. In general both are improving over the internet, but IPv6 6to4 failure is still at 9% in 2015 and IPv6 unicast failure is at 2%. This can be compared to IPv4 failure at .2%. When the connection is actually made, IPv6 performs as fast as IPv4. Geoff’s conclusion, of course, is that the failure rate of IPv6 is higher than the failure rate of IPv4.

Dario Rossi of Telecom ParisTech presented on the performance impact of Carrier Grade Network Address Translation (CGN).

The purpose of CGN is to allow many users to share one public IPv4 address. The finding presented by Dario was that CGN has no impact on the performance – that having one user with one IPv4 address is no more effective than having many users leveraging one IPv4 address using CGN.

The third presentation of interest was George Michaelson’s presentation on “Drilling Down into Networks from the APNIC Data.” I am not sure that this was George’s intent, but my takeaway was that there is a way to implement IPv6 called 464xlat, which is not supported by Apple iOS. This is important because several carriers have implemented IPv6 using 464xlat, which means that they won’t be able to carry Apple iphones over IPv6 until either Apple supports 464xlat, or until the carrier changes the way IPv6 is implemented. In George’s presentation the carriers that appear to have implemented IPv6 counter to Apple’s iOS strategy include T-Mobile, South Korea Telecom, AT&T Mobility, Sprint, and Orange Poland. George did include in his analysis that, in the right circumstances, 464xlat can boost IPv6 capability and reduce pressure on CGN/NAT. However, it would seem like 464xlat is an impediment to full conversion to IPv6 that needs to be corrected in the longer term.

Lastly, each Regional Internet Registry provided an update.

• AFRINIC has 2.2 /8’s left (37 million IPv4 addresses)
• ARIN Region has seen 188 companies added to the waiting list for returned blocks from the IANA. ARIN received a /14 from the IANA in September which went to 14 companies. ARIN is also expected to receive a /11 again in February, which will go to companies on a first in first out basis. ARIN has now completed 362 8.3 transfers to Specified Recipients and 181 8.4 inter-RIR transfers to or from APNIC.
• The APNIC presentation focused on trends, not raw numbers, and showed that while APNIC membership continued to grow in 2015, it appears the number of IPv4 transfers has tapered off from 2014.
• LACNIC has 2.1 million IPv4 addresses left under its soft landing policy where a company can get a /22 every 6 months. Inter-RIR transfers have been abandoned for now and transfers within LACNIC are not yet allowed.
• RIPE has a little more than one /8 remaining and there is much discussion about whether to change the methodology for the disposition of the remaining free pool, where an entity can receive a free /22 once.

My conclusion from RIPE 71?

There is a lot of contradictory information about the success of IPv6 in real life. It appears to me that strategies for stretching IPv4, such as CGN, are currently more effective than IPv6 standalone. This is based on IPv6 failure rates and applications not being supported over IPv6, and that the RIR’s are seeing strong evidence of continued IPv4 use in late 2015.