Pricing Networks with Selfish Routing

Abstract

We study the negative consequences of selfish behavior in networks and economic means of influencing such behavior. We focus on a simple model of selfish routing, defined by Wardrop and first studied from a theoretical computer science perspective by Roughgarden and Tardos. In this model, we are given a directed network in which each edge possesses a latency function, describing the common latency (delay) experienced by all traffic on the edge as a function of the edge congestion. There is a fixed amount of traffic wishing to travel from a source vertex s to a sink vertex t, and we assume that the traffic comprises a very large population of users, so that the actions of a single individual have negligible effect on network congestion. A common way to measure the quality of an assignment of traffic to s-t paths is by the sum of all travel times the total latency. We assume that each network user acts selfishly and routes itself on a minimum-latency path, given the network congestion due to the other users. In general such a "selfishly motivated" assignment of traffic to paths (a Nash equilibrium) does not minimize the total latency; put differently, the outcome of selfish behavior can be improved upon with coordination.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jul 28, 2003
Accession Number
ADA637950

Entities

People

  • Richard Cole
  • Tim Roughgarden
  • Yevgeniy Dodis

Organizations

  • Cornell University

Tags

Communities of Interest

  • Energy and Power Technologies

DTIC Thesaurus Topics

  • Algorithms
  • Computational Complexity
  • Computer Science
  • Computers
  • Congestion
  • Cost Reductions
  • Costs
  • Distribution Functions
  • Electronic Mail
  • Heterogeneous Networks
  • Money
  • Network Protocols
  • Networks
  • New York
  • Polynomials
  • Taxes
  • Theoretical Computer Science

Fields of Study

  • Computer science

Readers

  • Computer Networking
  • Economics
  • Graph Algorithms and Convex Optimization.