Revisiting the internet hourglass: core strength vs. middle-age spread

Abstract

The threat of commoditization poses a real challenge for service providers. Offering only a "plain vanilla" IP packet delivery service limits the options for competitive differentiation. Conversely, embedding additional functionality in the network carries a number of risks -- decreased robustness and increased complexity, for example. The key to addressing this challenge is the careful selection of appropriate functionality to embed in the network. Functions should be added to the network only when they offer value to a wide range of applications, and they should not inhibit the correct operation of applications that do not need them. This talk addresses the question of how novel, useful functions might be embedded "inside" the network, and how best to evaluate candidate functions for inclusion. For device designers, it is important to understand not only what functions are needed in the network today, but also which ones might provide the most benefit in the future. Because of the uncertainly about exactly what future networks will be expected to do, functions that are selected for inclusion in network devices must be as general as possible, and they should not interfere with the correct operation of the network when they are not needed. Some functions are best implemented as an overlay, leaving the essential network-layer functionality unaffected, while others will need assistance from the fast-path forwarding hardware. We will consider examples of various functions that have been or could be added to "core" networks, aiming to understand the tradeoffs both among different functions to add and among different implementation approaches.