Do we need computing in the network?

The commodisation of basic networking functionality is reigniting the debate over whether the intelligence should reside within the network infrastructure, or simply use the network as a pipe.


The 1990s saw the commoditisation of LAN technology. Today we are seeing the beginning of commoditisation of the IP routing function in the intranet, with more on the way as new players enter the network element market.

Network functions that once could provide vendors and providers with a strong revenue stream eventually may become near-loss leaders. For example, 75 years ago it cost $7 a minute to make a telephone call from New York to Las Vegas; today such a call can cost as little as 2 cents on the traditional network infrastructure and even less on the Internet -- an annual compound erosion of 7.5 percent. Twenty years ago a LAN network interface card cost more than $1,000; today it typically costs less than $10 -- an annual erosion of about 20 percent. With this kind of erosion, a $10,000 router will cost $100 before many of us retire. Forward-looking vendors are keenly aware of these technology trends.

To address this liability, router vendors are beginning to look into adding job functions such as network-based computing to their products. With what some vendors call application-oriented or application-aware networks, functions such as proxying, content-level inspection and advanced voice-data convergence are being advocated. In-network processing and data analysis have been proposed for large sensor networks; however, commercial deployments of sensor networks have by and large not yet implemented these in-network features.

Proponents view this trend toward in-network computing as the ultimate form of convergence: the network function is more holistically integrated with the application, particularly where QoS requirements exist. Others see this trend as a mechanism to move to a service-oriented architecture, where applications are expressed in terms of functional catalogues and network functions are seen as addressable, executable blocks.

These recent forays into network-computing hybridisation are reminiscent of the telephone carriers' efforts in the late 1980s to build an Advanced Intelligent Network that would provide value-added capabilities in the transport fabric. They also remind us of the tug-of-war that has existed between the traditional carriers, who want to maximise the intelligence and value proposition of the network, and the IP data people who want the network to have very thin functionality.

In trying to make the network intelligent, some inherent generality is lost as functions specific to targeted tasks and applications are overlaid. Nonetheless, it is interesting to note that the same people who once argued against network intelligence advocate this idea now that their revenues are eroding.

Should we expect to see the emergence of application-oriented networks soon? Probably not in the short term. However, commoditisation of the routing function will pick up speed and routers will become easier to configure and administer reliably. If the router's command-line interface is not modernised, and the cost of router configuration and administration is not reduced, some new, more cost-effective technology will eclipse it at some point in the next decade.

Daniel Minoli is an adjunct professor in the Stevens Institute of Technology's graduate school.

"Recommended For You"

Cellular in the last mile - hidden problems Bitdefender security appliance for home networks seeks to replace end-point antivirus