For decades, it's been an axiom of service provider network design that "IP networks" means "router networks." Like many of the axioms of networking, though, this one is showing signs of crumbling.
At the core of the changes is the transformation in IP services, from basic Internet to broadband premium IP services for both the consumer and the enterprise. But technology changes are creating their own momentum as well.
When the Internet became popular in the 1990s, services were dial-up, best-effort connections. There were thousands of Internet Service Providers (ISPs), their ranks inflated by the effects of the venture capital bubble. Worldwide IP infrastructure was a consortium of larger and smaller providers, with both public and private peering between them. IP addresses often moved from provider to provider as companies changed ISPs, and the cost of bandwidth encouraged a hierarchical aggregation of traffic toward a dense core. Routers, devices that provided any-to-any connectivity and connection management for IP addresses, were essential.
This decade, has already seen a tremendous consolidation among Internet providers, however, partly due to the collapse of the bubble and partly due to the broadband revolution in consumer access. Consumer broadband is a big-capital-investment market that only large players can hope to enter. Because of that, the majority of customers were concentrated onto a smaller collection of ISPs.
The consumer interest in content has also been changing the way traffic flowed. The demands of content-rich websites expanded the use of caching at the edge to reduce transit cost and delay. This has short-circuited much traffic at the inside edge of the access network, creating more bandwidth growth in the metro area to get to the cache points, but reducing the growth in the core. Hosting IPTV content, which is almost totally a metro process, widens the growth gap between metro and core capacity.
The metro/core gap is important because most metro networks have been dominated by optical, ATM and Ethernet investment rather than IP. The DSL Forum architecture for metro broadband uses tunnels from the user to a Broadband Remote Access Server (BRAS) that serves as the IP network onramp. These tunnels not only don't require IP under them, they work best with simple Ethernet/optical infrastructure for cost and security reasons. Thus, as metro growth outpaces core growth, optical and Ethernet switching outpace routing growth. By the time significant IPTV deployments occur to further explode metro-only traffic, likely by 2010, most IP traffic won't really see much router handling because it will stop inside the metro network.
Switching in the Core
Even in the core, there are already signs of a transformation to switching. Level 3 has been touting its switched core for several years now, and as optical bandwidth costs decline, it makes more sense than ever to build networks with fast simple switches (Level 2 or 3) in the core and to limit routing functions to network onramps. Today, two factors are driving service providers to look harder at switching even in the core -- legacy convergence and portals.
According to service providers worldwide, convergence of legacy frame relay, ATM and TDM services onto IP has resulted in a huge increase—four to five times—in the number of customer support incidents. Some of this is attributed by the providers to the adaptive nature of IP routing, which makes it difficult to control premium paths and to write precise service level agreements (SLAs). BT and other providers have expressed the desire to use advanced Ethernet standards (Provider Backbone Bridging and Provider Backbone Transport) to create Ethernet switched substrates to their IP networks. Legacy services would then converge not onto IP but downward onto Ethernet and switching.
Portals May Influence Infrastructure Investment
Portal vendors like Google are driving another trend that has only begun to be felt is driven. To understand how it impacts switching/routing, imagine an Internet where all of the content is hosted by a single company from a single data center. That company could simply run a fiber connection to the access networks of all of the major broadband providers and reach 80% or more of customers without any of their traffic flowing on "the Internet" at all. Furthermore, only traffic between the users and their datacenter would flow on the fiber.
This isn't as farfetched as it might seem. The DSL Forum (TR-059) recognizes two types of service connections to a DSL network—the traditional "ISP" and an Application Service Provider (ASP). The latter simply provides access to ASP resources and does not assign IP addresses or carry user traffic. Portal players like Google can use the ASP attachment to pump their content directly to the access provider. How much routing is required for that point-to-point connection?
Universal any-to-any connectivity requires IP addressing and routing somewhere, so routers will not disappear. But because profitable traffic doesn't flow any-to-any, and because infrastructure investment is made where return is the highest, it is very likely that more and more IP will be served over switching in the future.
About the Author: Tom Nolle is president of CIMI Corporation, a strategic consulting firm specializing in telecommunications and data communications since 1982. He is a member of the IEEE, ACM, Telemanagement Forum, and the IPsphere Forum, and the publisher of Netwatcher, a journal in advanced telecommunications strategy issues. Tom is actively involved in LAN, MAN and WAN issues for both enterprises and service providers and also provides technical consultation to equipment vendors on standards, markets and emerging technologies. Check out his complete networking blog.