Changes in Internet traffic delivery could lead to the resurgence of dedicated networks by 2011-2012, just as Internet demand exceeds capacity at the edges of the network and IPv4 addresses are depleted, according to a new Nemertes Research study, Internet Interrupted: Why Architectural Limitations Will Fracture the Net.
Changes in the Internet's structure create other issues because the fragmentation that Nemertes Research sees is really a result of the Internet growing up and becoming a market-driven economy, according to Nemertes analyst Ted Ritter, who cites Google's investment in undersea Pacific cable as an example of a content company financing a traditional service provider venture for competitive advantage.
The Nemertes report serves as a warning, however, that the Internet community as a whole must focus on making sure that core BGP routing tables aren't overloaded as traffic grows and individual IP addresses proliferate. In terms of the growing number of IP-enabled devices, even though a network upgrade to IPv6 can supply millions of new addresses as IPv4 addresses run out, compatibility issues must be addressed.
In its 2007 Internet demand and capacity study, Nemertes modeled Internet and IP infrastructure and projected traffic growth. This year, analysts revisited last year's finding and looked at logical constraints on the Internet, as well.
"We think the Internet is getting flatter as content providers are moving their content closer to user eyeballs," Ritter said. By "flatter," he explained, he means that the old view of the Internet had a hierarchy with global tier 1 backbone providers connected to regional tier 2 providers, which then connected to local providers. Instead, companies like Google, Limelight and Akamai don't want to have to transport content through so many network hops.
Service provider investment is currently focused on building up metro networks for content delivery, yet other network changes are being driven by the need to guarantee quality content delivery. IP traffic routed through public peering points isn't growing as fast as many projections suggest, which may mean that more traffic is being routed through private network connections, but is difficult to measure.
"Respected researchers who monitor public peering points can't see traffic going through private peering points or traffic that is being provided through parallel, flatter high-speed lines that content-distribution networks and providers are putting in," Ritter said. The Internet's subtle "shape shifting" has huge implications for Net neutrality and the future working of the Internet. If the Internet is moving to a new structure, he said, the ability to monitor and regulate it also needs to change.
IPv6 isn't Internet's silver bullet
When analyzing logical Internet issues in this year's study, Nemertes also focused on challenges presented by the exhaustion of IPv4 addresses, which it projects will run out by 2012 as a result of the exponential growth of Internet-connected devices and machine-to machine communications. Service providers have largely upgraded their networks to IPv6, but only 1% of enterprise IT decision-makers participating in a Nemertes benchmark study have deployed IPv6.
A new market could grow out of what has been an essentially free resource – reselling and reallocating IP addresses, Ritter said. When addresses are reused, tremendous stress is placed on the router infrastructure.
Many people believe that changing to IPv6 is the simple answer, but -- according to Ritter -- Nemertes believes this is the biggest problem it looked at this year. "IPv6 can give an IP address to almost every molecule on Earth," he said. "But you can't get there from here." There are architectural problems that can be solved, but the transition will be difficult.
Router infrastructure isn't designed to route individual addresses, but blocks of addresses. "Having an IPv4 address talk to an IPv6 address is not a trivial process," Ritter said. "It's going to require specialized gateways in service provider clouds. It requires convoluted efforts."
Even though translation problems can be solved, core routers handle millions of packets per second. "For them to stop and look at a packet and do translations and lookups is a huge performance hit," Ritter said, adding that the gateway boxes have to be developed to perform at scale.
The takeaway from Nemertes is that IPv6 isn't an easy answer and that service providers and enterprises will probably live with IPv4 much longer than expected, requiring providers to do network address translation in the cloud.
"I'd love for us to be wrong, and if someone could just throw a switch and go to IPv6 without huge implications, it would be great," Ritter said. "But this address depletion and routing challenge has to be solved globally."