IPv4 and IPv6 will coexist on the Internet for decades, creating the need for additional transition mechanisms because the dual-stack model won’t solve all of the emerging problems for network operators.
That was the consensus view of a panel of experts who discussed IPv6 operations and transitional Issues at the Internet Architecture Board (IAB) technical plenary held 8 November in Beijing.
“We’re going to have a very long transitional coexistence period,” said Danny McPherson, chief security officer at VeriSign, who moderated the panel discussion. “There’s a lot of work going on still for . . . some of the strategies for IPv4-only devices to speak to IPv6 networks where you don’t have dual stack as an option.”
Latency is a key issue that the cable company Comcast has run into during its ongoing public trials of several IPv4-to-IPv6 transition mechanisms. Comcast began its IPv6 deployment five years ago, and its network is largely dual stack, along with its back office functions and access network.
Comcast noticed that it had a large volume of tunnelled 6to4 traffic on its network and that these end-users were experiencing too much latency. So the company deployed its own 6to4 relays based on an open source Linux platform, and that has improved the performance of 6to4 traffic by 50 percent or better.
Comcast also tested the 6rd encapsulation mechanisms and found that they performed better than 6to4. It found the 6rd relays to be “extremely simple to deploy,” said John Brzozowski, chief architect for the IPv6 programme at Comcast. “If your access networks cannot support native IPv6, [6rd] definitely feels like something that you should look at.”
John pointed out that network operators will see their costs rise as they deploy additional IPv6 transition mechanisms. “The more you have to put out there for subscribers, the more the investment will be for you,” he added.
One issue that is driving up the cost of Comcast’s public IPv6 trials is its need to manually configure all of the customer premises equipment (CPE). “We used just shy of 300-some odd devices,” John said. “We had to manually configure each and every one of them before we shipped them out to the trial users. That’s clearly not going to be scalable long term.”
Comcast said dual stack will offer its subscribers the best overall experience because it has direct end-to-end routing without translation, tunnelling, or encapsulation. In the future, Comcast hopes to be able to purchase CPE that supports native IPv6, but it still faces the challenge of dealing with older CPE that isn’t upgradable to IPv6 and must be replaced.
“Operators should go with what’s available to them now,” John urged. “Get started. Really, don’t wait until it’s perfect.”
Matsuzaki Yoshinobu of Internet Initiative Japan said he has run into several unexpected implementation issues with routers handling IPv6 traffic. Some routers only support lower prefixes while others have trouble sending bigger IPv6 packets. Packet filtering for IPv6 also can be problematic.
Another issue Matsuzaki noted is poor user experience, such as a lack of IPv6 connectivity in some countries and the poor performance of IPv6-over-IPv4 tunnelling offered by some ISPs. Other problems are broken discovery and the need for link-local addressing.
China Telecom is trying to move rapidly to IPv6 because it needs about 30 million IP addresses in 2011 alone to support its rapidly increasing subscriber base for broadband, IP television, and other services. But it has only 10 million IP addresses, leaving a gap of 20 million IP addresses.
“For a lot of new services, new applications, we still need several billion new addresses in the future five years,” said Huiling Zhao of China Telecom.
Huiling said China Telecom is exploring four ways to meet this demand for IP addresses: reusing existing IPv4 addresses, using private IPv4 addresses, purchasing additional IPv4 addresses, and deploying IPv6. “We think IPv6 is the best solution in the future,” she added.
Huiling said there are problems with each of the existing IPv6 transition mechanisms. Dual stack requires a dedicated IPv4 address for each user, and it also has performance problems. NAT444 lacks carrier-grade performance and is difficult to deploy on a large scale. DS-Lite requires that home gateways be upgraded. Currently, China Telecom is studying additional mechanisms including NAT64, IVI/DIVI, and 6rd.
“Perhaps we finally need a cocktail method combining several tunnelling and protocol transition methods in order to meet our market requirement,” she said, adding that one possibility is combining dual stack with private IPv4 addressing.
Xiaodong Lee of China Internet Network Information Center (CNNIC) said that the number of registered IPv6 addresses and the amount of IPv6 traffic are very small in China despite the presence of a large IPv6 network. In fact, Xiaodong said there are many more IPv6 users in Europe and America than in China.
“There is no strong requirement for . . . users to use IPv6,” Xiaodong said. “The reason for this is because the user, they don’t care what is IPv4 or IPv6. They only care about the application. . . So the killer issue is applications.”
Bill Huang from China Mobile questioned whether the dual-stack model will work well enough to support the migration to IPv6 at the same time that the company’s network is projected to grow as much as 100-fold over the next five years. He said most dual-stack configurations default to IPv4. Instead, he favors the creation of new traffic-steering protocols that will translate or tunnel traffic from an IPv4 network to an IPv6 network.
“The result is that we will be able to see more and more traffic being steered towards a pure IPv6 network,” Bill said. “If we equipped a new generation of terminals with these types of technology . . . then by default the traffic will be steered.”
Jari Arkko of Ericsson concluded the panel with findings from his research of IPv6-only networks. He said several applications, including browsing, email, software updates, and streaming music work very well, while others, such as gaming and Skype, do not.
“I think we should still keep on recommending dual stack as the preferred mode. It has the least amount of problems,” Jari summed up. “We can recommend IPv6 only as well for early adopters and mobile networks.”
This article was posted on 16 March 2011