Article,
Internet Congestion: A Laboratory Experiment
Proc. SigComm'04, (2004)
Abstract
Human players and automated players (bots) interact in real time in a congested network. A player's revenue is proportional to the number of successful ``downloads'' and his cost is proportional to his total waiting time. Congestion arises because waiting time is an increasing random function of the number of uncompleted download attempts by all players. Surprisingly, some human players earn considerably higher profits than bots. Bots are better able to exploit periods of excess capacity, but they create endogenous trends in congestion that human players are better able to exploit. Nash equilibrium does a good job of predicting the impact of network capacity and noise amplitude. Overall efficiency is quite low, however, and players overdissipate potential rents, i.e., earn lower profits than in Nash equilibrium.
