INET Denver considers Internet life without IPv4 addresses

This was originally posted on Arstechnica

After Asia and Europe, North America is next in line to run out of IP addresses.

On Wednesday, the Internet Society, the Rocky Mountain IPv6 Task Force, and others organized “INET Denver: IPv4 Exhaustion and the Path to IPv6.” After seeing the same scene play out in Asia in 2011 and in Europe last year, the North American Internet industry seems well-aware that they’re next in line to see their access to (almost) free and (fairly) plentiful IP addresses dry up. At this time, the American Registry for Internet Numbers (ARIN) has 2.44 blocks of 16.78 million IPv4 addresses left. This amount is predicted to last until April 2014.

After some introductory remarks, Geoff Huston, who works at ARIN’s Asian counterpart APNIC, got the ball rolling with an overview of the state of IPv6. (The slides he used were very similar to these from November (PDF).) Huston argued that the telecoms industry has “a rich history of making very poor technology choices.” This is the same industry that gave us ISDN and ATM—after all, that Internet thing and the packet switching that it’s based on would never amount to anything. Huston went on to explain the environment in which IPv6 was created: IPv6 came about in order to create additional IP addresses to fuel the Internet once the some four billion that the existing Internet Protocol version 4 (IPv4) allows for are used up.

Running out of IPv4 addresses was predicted in some fashion as far back as two decades ago. But Huston showed how despite this, the IPv4 piggy bank now looks extremely empty and it still managed to catch much of the industry by surprise. A big part of the problem was that until now, new technologies always allowed the network operators to save money. With IPv6, that’s not the case. Huston talked about several transition strategies, but he doesn’t find any that will provide relief before the shortage of IPv4 addresses becomes acute. However, there are ways to slice and dice IPv4 addresses so existing and new groups of users can continue to be connected to the IPv4 Internet. But the network operators that deploy these technologies now have an incentive to protect their investments in expensive equipment.

According to Huston, at some point during the next five years we have to make a choice. We can go down the path of Carrier Grade NATs (CGNs) and put an entire neighborhood behind a single shared IPv4 address. Or, we can bite the bullet and upgrade or replace all those devices (mostly in the last mile infrastructure) that are keeping us from moving towards IPv6. “And it’s not yet clear which path the Internet will take!”

However, it seems that expecting the entire Internet—or even just the network operators providing access to it—to do the same thing is an exercise in futility. Surely there will be many CGNs deployed in the next five years, while the number of people with IPv6 connectivity will keep going up simultaneously. That crossroads will remain a busy place as people have to keep coming back to it and switch paths on an ongoing basis.

Huston finished with a few observations, including his belief that addresses should be used not hoarded. This brings us to the “Evaluation of Current Transfer Market” panel discussion. We’ve covered address trading before. Since then, it has become possible to transfer IPv4 addresses between ARIN and APNIC. A few blocks have been transferred from North American holders to organizations in Asia and the Pacific. The inter-region address trading can only happen when the two regions have compatible policies, which is currently not the case between ARIN and the RIPE NCC in Europe. That may or may not change, depending on the policy proposals that will be adopted by the RIPE NCC in the near future.

Two representatives from companies that facilitate address transfers were also part of the panel, and both advocated for fewer rules getting in the way of address transfers. (“IPv4 isn’t exhausted, it’s not even tired!”) Interestingly, the five Regional Internet Registries aren’t in charge of their own policies: these are created through a community-based policy development mechanism. Unfortunately, any adverse results, such as the creation of lots of small address blocks, affect the entire world.

After all the address trading discussions, Time Warner Cable’s Lee Howard talked about the total cost of ownership (TCO) of CGN and IPv6. CGNs aren’t cheap, and they break some applications leading to support calls. But these two costs are dwarfed by another undesired result: if their favorite application no longer works, some users may terminate their contract. As a result, the costs of deploying CGN are dominated by the lost revenue of customers that leave because CGN breaks their applications. Running IPv6 alongside IPv4 (“dual stack”) requires new software and, for many users, a new cable or ADSL modem. The addition of IPv6 doesn’t break anything, so there’s no lost revenue. But again, it also doesn’t solve the shortage of IPv4 addresses in the short term.

Last but not least, there’s the option of buying IPv4 addresses. Howard calculated that there’s roughly half a billion abandoned or unused IPv4 addresses. These may enter the market with prices between $9 and $12 per address. Another half a billion may be underutilized and become available for between $9 and $16. Considering the costs of CGN deployment and/or dual stack, being able to buy IPv4 addresses for these prices would be attractive for ISPs. However, if that starts happening on a large scale, the unused and underutilized IPv4 addresses will still be used up by 2017. After that, IPv4 addresses would have to be bought for as much as $100 or more per address. That is simply untenable for ISPs.

We spoke with Richard Jimmerson, now IPv6 expert at the Internet Society but formerly at ARIN, who thinks we’ll see the results of both good and bad planning in upcoming years. “Businesses that planned well will continue to grow without hiccups. The ones that planned badly will see hiccups with ongoing business new services. However, it’s unknown how much of this we’ll see on the outside.” Unlike Huston, Jimmerson thinks we’ve already passed the fork in the road. “I believe the industry will do the right thing and deploy IPv6. If you’d asked five years ago, you’d have gotten a completely different answer. But people are onboard with IPv6 now.”

Even if IPv4 is going to be around for many years, at some point, the old protocol may only be a shadow of its former self. “There are going to be reasons to do IPv6 earlier than expected,” Jimmerson said. “For instance, IPv4 may be so spaghetti-strapped that performance is worse than IPv6.” Spaghetti-strapped in this sense means being overprovisioned, so too many users have to share IPv4 resources with too much NAT and too many Band-Aids to be useful. At that point, IPv4 is nothing more than a legacy technology. But, Jimmerson noted, “I wish more energy went into IPv6 and less into making CGN work.”

Unlike the last two years, the Internet Society won’t be organizing an event like World IPv6 Day orWorld IPv6 Launch in 2013. As for the future, “it’s up to industry players” to decide what possible future IPv6-related Web-wide events will entail.

If anything, INET Denver demonstrated once again that there are no easy answers when it comes to running out of limited resources. But—we can’t avoid the question forever.