The Internet as it exists today may come to an end in 15 to 20 years because of flawed routing architecture; but, in the meantime, plenty of work must be done to keep what is already deployed from splintering.
That was the general agreement (with some vocal dissenters) at FutureNet's annual "Beer and Pizza" shootout, as a distinguished panel of Internet architects and thinkers debated the urgency of -- and the possible solutions to -- two fundamental problems: IP address depletion and the scalability of the Internet's routing capabilities. Many attendees disagreed about whether the problems should be dealt with together or separately.
The latter problem occurs because of the hierarchical way service providers route IP addresses. While such methods have worked well as billions have logged online, the combination of billions more IP-enabled devices, coupled with much more frequent handovers as mobility increases, puts more strain on the scalability of the Internet than its original architects imagined it would face as a largely academic enterprise.
One prominent voice, among many, on the panel was John Curran, acting president and CEO of the American Registry for Internet Numbers (ARIN), which is the regional Internet registry charged with doling out IP addresses to the United States, Canada and the Caribbean, and the North Atlantic Islands.
Curran was also formerly the CTO of XO Communications and CTO of BBN Technologies.
He insisted that solving both problems together is impossible because address depletion is about to become an immediate problem that will reach a crisis level in two and a half years. It has a generally accepted solution in IPv6. The flawed routing architecture of the Internet will not come to a head for another 15 to 20 years, and there is not yet general agreement on how to solve it.
"They're saying going to v6 doesn't solve the long-term problem, and it doesn't," Curran said in an interview after the debate. "It solves the short-term problem. The Pouzin Society and GENI initiative at the National Science Foundation are looking at the next, next problem, which is decades down, but worth doing."
Unfortunately, history has taught us that proactive approaches to fixing the Internet often founder, Curran said. The IPv4 address crisis is a perfect case in point. IPv6, its successor protocol, was finalized 15 years ago, and for all the innovative ideas proposed for it, such as improved security, the only problem it really solves is the addressing crisis.
Groups push for revised network infrastructure
Meanwhile, the work of the Pouzin Society and GENI is at this point extremely academic, without even any working code or proof-of-concept work finished.
In fact, the Pouzin Society held only its first meeting at FutureNet this week.
The society, founded by John Day, is pushing a revised network infrastructure based partially on his book Patterns in Network Architecture, which advocates using Inter-Process Communications (IPC) to eliminate the routing scaling problems the Internet is suffering. These ideas, which Curran said have some merit, will simply require too much time to implement before the more immediate IP addressing crisis comes to a head.
Indeed, while the IPv6 transition took pains to be as backward-compatible as possible, the Pouzin Society and GENI are both proposing more fundamental re-workings of the Internet to address not only the routing issue but other fundamental lessons learned since the launch of the Internet.
A great vision, Curran said, but one totally incapable of being made reality within the next two years.
"You can't kill two birds with one stone because you need to have a stone within the next two years, and the other stone is lava," he said. "We have to deal with a pressing operational issue, and we only have one protocol choice to make."
For the long term, however, a "clean slate" approach is needed, Curran said -- exactly the sort of re-architecture the Pouzin Society and NSF are working on.
"There is a valid reason to study these things," he said. "If you presume that there are billions of devices with IPv6 addresses, all dynamically connected to the network and connected to multiple networks -- the hotel wireless, your WAN provider, your cellular company, a local show network. That creates multiple paths for hundreds of different systems in your laptop, all of which have IP addresses. Moving between ... the networks could generate thousands of updates dynamically. That's the world of the future. Multiply that by billions of devices and the routing problem is huge."
Huge -- but, in Curran's estimation, decades off.
"How will the network fare over the next two decades while we wait for this to settle?" he asked. "Well, the carriers ... will keep the network running. Will there be side effects? Yes -- maybe you don't get to carry your address around; maybe you do have to renumber; and maybe it's not dynamic so that when you switch from wireless network A to wireless network B, all your sessions drop. [Carriers] will figure out how to damp down the routing to keep it stable."
Trying to rush any solution other than IPv6 in the meantime will only end in misery, however.
"It's taken us 15 years to get IPv6 deployed in my laptop, your laptop, cell phones, carrier equipment, the equipment on the floor here ... and that was after we finished the specifications," Curran said. "Anything with a clean design, we're probably looking at the 20 to 25 year time frame before we're installing it in a router near you."
Redeploying IPv4 addresses has major downside
Curran also responded to those critics, including those at FutureNet, who asked why the massive number of unused IPv4 addresses allocated to corporations or nations could not be redeployed, delaying the IPv6 transition and possibly allowing enough time for the routing crisis to be addressed.
He said the fragmentation that would result from splitting up IPv4 routing addresses would only exacerbate the routing problem while simply delaying the inevitable IPv6 transition by a few years.
"If you don't go to v6 as an option, if you have to fill your routing table by filling it with lots of little pieces [of v4], that's going to be a problem," he said. "[With dual v4/v6 setups] there's a little bit of double routing, because you have to deal with two tables, but that's better than the fragmentation that will occur if you don't go."
But, at the end of the day, it comes down to whatever keeps the networks running, Curran said.
"The side effects aren't pure from a network purist's perspective, but rest assured that if it comes down to keeping an operational network or keeping a pure architecture, we'll do the same thing we did before and throw out the pure architecture," he said. "So the routing system will be managed to keep things under control, as long as we go to v6. If we don't go to v6, that might not be possible."