Network Dynamics

The Internet as a Complex System: Scaling, Complexity and Control

Kihong Park, Computer Science, Purdue University
Walter Willinger , AT&T Research

The Internet is a "complex system" comprised of many interacting elements connected by various network media whose behavior is governed by control programs called network protocols. The latter encapsulate a wide spectrum of functionalities-e.g., routing, congestion control, error control, quality of service control-and, in most cases, are decentralized. Two recent developments pose new challenges to the effective design, analysis, and control of the Internet: rapid explosion of scale and demand for quality of service. Scalability looms as an imposing problem to the control algorithms of the past, and the provisioning of heterogenous quality of service (QoS) requires the close, unprecedented cooperation of network elements. The Internet is, perhaps, the largest manmade many-body system whose potential complexity is only beginning to surface.

For example, the discovery of self-similarity in network traffic is an instance of an unsuspected, "emergent" system trait whose impact on network performance and control has transcended its phenomenological roots. Chaotic dynamics exhibited by TCP and the multifractal property of wide area IP traffic are conjectured to stem from the feedback nature of Internet traffic control. Power-law scaling observed in network topology points toward invariance structure manifesting both in time and space. The influence of user behavior-the typical driver of network applications-on network behavior is another important factor. At one level, the network system, when habited by selfish users characteristic of the Internet, takes on the traits of a noncooperative game, and its stability and efficiency properties are integral to understanding the overall system.

This upcoming meeting will bring together leading researchers in networking whose works have touched upon the "Internet as a complex systems" theme, joined by a select group of physicists, mathematicians, and economists with expertise in statistical mechanics, dynamical systems theory, and game theory. The meeting is expected to (a) take stock of past accomplishments and evaluate the state-of-the-art in networking with respect to scalability, dynamics, and performance; (b) discuss the extent to which understanding and engineering the future Internet will be different from the past, and how it can benefit from interdisciplinary approaches; and (c) identify a set of grand challenges and disseminate promising avenues of attack. In the current, specialized network research environment, it is not easy to step back from one's existing commitments and explore dramatically new ideas, or be exposed to varied points-of-view. It is hoped that the meeting, set in an informal atmosphere, will foster stimulating interaction, exploration, and exchange of ideas.