Lack of IPv6 traffic stats makes judging progress difficult
The Internet is poised to undergo the biggest upgrade in its 40-year history: from the current version of the Internet Protocol known as IPv4 to a new version dubbed IPv6, which offers an expanded addressing scheme for supporting new users and devices.
However, it will be difficult for Internet policymakers, engineers and the user community at large to tell how the upgrade to IPv6 is progressing because no one has accurate or comprehensive statistics about how much Internet traffic is IPv6 vs. IPv4.
The issue of IPv6 traffic measurement is timely given that the Internet engineering community is preparing for its biggest trial of IPv6: World IPv6 Day on June 8. So far, 225 website operators -- including Google, Yahoo and Facebook -- have agreed to participate in the event by serving up their content via IPv6 for 24 hours.
Without accurate IPv6 traffic statistics, neither the sponsors nor the participants of World IPv6 Day will be able to tell for sure how much IPv6 traffic is sent over the Internet on June 8 or how much difference the event has on IPv6 traffic volumes afterward.
"Being able to measure IPv4 vs. IPv6 is very important," says John Brzozowski, distinguished engineer and chief architect for IPv6 at Comcast, which has deployed an emerging tool called NetFlow 9 in parts of its network to measure IPv4 and IPv6 traffic volumes. "We as a community need to be able to measure our progress and success. IPv4 vs. IPv6 traffic is one of many important metrics."
IPv6 traffic data "is useful in understanding the trend of adoption, but I don't think it's critical for a carrier to understand that they have to properly support their customers," says Dave Siegel, vice president of IP Services Product Management at Global Crossing. Siegel adds that "it's most important to have tools available to troubleshoot IPv6 issues."
IPv6 traffic volumes are likely to remain a hot topic as the pressure intensifies for network operators to deploy IPv6, a 13-year-old standard whose primary advantage over IPv4 is an expanded addressing scheme. While IPv4 uses 32-bit addresses and can support 4.3 billion devices connected directly to the Internet, IPv6 uses 128-bit addresses and can connect up a virtually unlimited number of devices: 2 to the 128th power.
The Internet needs IPv6 because it is running out of IPv4 address space. The free pool of unassigned IPv4 addresses expired in February, and in April the Asia Pacific region ran out of all but a few IPv4 addresses being held in reserve for startups. The American Registry for Internet Numbers (ARIN), which doles out IP addresses to network operators in North America, says it will deplete its supply of IPv4 addresses this fall.
DETAILS: Asia out of IPv4 addresses
But as necessary as IPv6 seems, there is a major stumbling block to its deployment: It's not backward compatible with IPv4. That means website operators have to upgrade their network equipment and software to support IPv6 traffic. So far, most have been unwilling to do so because IPv6 traffic has been so scarce.
One of the only regular surveys of Internet traffic is compiled by Arbor Networks, which recently reported that IPv6 represented less than 0.2% of all Internet traffic. Arbor said IPv6 traffic -- both tunneled and native -- had declined 12% in the last six months, even as momentum for World IPv6 Day was building. Arbor gathered this data by surveying six carriers in North America and Europe.
Not everyone believes Arbor's assertion that IPv6 traffic is declining while so many website operators are preparing for World IPv6 Day.
"I did not see this [data], and I am also very surprised if it is accurate," says Russ Housley, chairman of the Internet Engineering Task Force, a standards body that created IPv6. "I am aware of many organizations preparing for World IPv6 Day."
Martin Levy, director of IPv6 strategy at Hurricane Electric, a Fremont, Calif., ISP that claims to have the world's most interconnected IPv6 backbone, says he is struggling with how the Arbor Networks data can be true given the network industry momentum behind IPv6. "Where Arbor measures is not where the predominant IPv6 usage is," Levy says.
BACKGROUND: The 6 biggest misconceptions about IPv6
What's preventing the Internet engineering community from coming up with more complete IPv6 traffic statistics is the fact that few ISPs or their hardware suppliers have deployed a network management tool called NetFlow 9. This industry standard export protocol sends data about traffic flows through the router to an external collection host so that the flow information can be analyzed. NetFlow 9 can be used to separate out and measure IPv6 and IPv4 traffic flows.
NTT America, a leading provider of IPv6 transit services, concedes that it doesn't measure the IPv6 traffic it is carrying separate from overall Internet traffic, so it doesn't know the rate at which its IPv6 traffic is growing.
"Most routers don't have support for counting IPv6 traffic separate from IPv4 traffic on their physical interface counters," explains Dorian Kim, vice president of IP engineering, Global IP Network at NTT America. "By and large, bits are bits to interfaces, and there's been no particular driver for us to track IPv4 and IPv6 separately, especially when measuring IPv6 in the network requires one to instrument NetFlow 9 collection of traffic from the routers."
Kim says NetFlow 9 is "the most likely way we'll be able to account for IPv6 traffic separately from IPv4 traffic. ...We are working with our equipment vendors to have NetFlow 9 export capabilities across all of our network, but that is something that'll happen over time."
NTT America isn't the only ISP that hasn't been able to deploy NetFlow 9. "I heard from a significant cable operator that they don't run NetFlow 9 and hence can't measure IPv6 flows," Levy says.
One reason most carriers don't have NetFlow 9 gathering statistics across their peering and edge routers is because it can cost millions of dollars to deploy.
"It's pretty expensive to do complete traffic measurement across your entire network," says Siegel, pointing out that Global Crossing has deployed NetFlow 9 in Europe. "It's kind of a luxury."
Siegel says Global Crossing will have NetFlow 9 deployed across a third of its network footprint by the end of the year. He says the carrier has run into trouble deploying NetFlow 9, including the discovery that its edge routers didn't have enough horsepower to collect data on every port, resulting in a limited view of traffic patterns. Global Crossing plans to use NetFlow 9 to create customized billing models rather than to monitor IPv6, which represents only 0.1% of its traffic in Europe.
"We only started making [traffic measurement] more of a priority lately as a way of analyzing customer profitability," Siegel says, adding that measuring IPv6 traffic flows is not a driver for the purchase. "The variability of how a customer is using the network plays an extremely large role in whether or not we see profit at any given price point. That's our primary purpose behind ponying up the cash to buy the [NetFlow 9] software and the hardware platform."
Some experts argue that measuring IPv6 traffic with NetFlow 9 is not the best way to determine the rate at which IPv6 is being adopted.
Traffic is too ephemeral, says Geoff Huston, chief scientist at APNIC, the regional Internet registry for the Asia Pacific Region. Huston prefers to measure how many end users have IPv6 enabled on their devices. He points out that this figure is as disappointingly low as the Arbor Networks IPv6 traffic data.
"When you look at end users and their predilection to use IPv6, the story is pretty bad," Huston says. "I see 0.2% of clients prefer to use IPv6 on a dual-stack environment. ... Since November, that number has risen by a little under 0.1%. Yes, that's a 50% jump. No, these are still tiny numbers and cannot be interpreted as a wave of IPv6 adoption."
Huston says end user IPv6 adoption rates are not rising faster because recent versions of the Windows and Mac operating systems prefer to use IPv4 in dual stack environments where IPv4 and IPv6 are running side-by-side, rather than using IPv6-based tunneling techniques such as 6to4 and Teredo.
"The reason why modern end host systems prefer to use IPv4 is that auto-tunneling using either 6to4 or Teredo is lousy. The experience simply is atrocious!" Huston says. "So when we measure the amount of end hosts that prefer to use IPv6 in a dual stack scenario we are in fact looking at old end host systems that are not being maintained in terms of software releases from the vendor and a small number of end hosts that have native IPv6 service from their service providers."
Huston says observers shouldn't expect IPv6 traffic volumes to rise significantly on or immediately following World IPv6 Day. That's because IPv6 traffic volumes will only rise for end users with native IPv6 service from their ISPs.
"How many folk will see a difference [on World IPv6 Day]?" Huston asks. "At most 0.3% of folk. It's hardly a shattering number."
Instead, Huston recommends that the Internet engineering community measure the success of World IPv6 Day in terms of reducing the fear that website operators have about the cost and hassle of upgrading to IPv6.
World IPv6 Day "is intended to turn the Internet, or at least the Web content delivery part of the network, into a laboratory for 24 hours [so we can] see what happens when a mono-stack world turns into a dual-stack world," Huston says. "In and of itself, it won't get the IPv6 ball rolling."
Read more about lan and wan in Network World's LAN & WAN section.