AbstractWe aim to reduce congestion through better signaling. That is, wepropose signals that a central agent with real-time data on the congestionacross a number of resources should send to the users, who compete forthe limited resources, in order to improve social welfare. We model aheterogeneous population of users and the interactions of this populationwith the central agent over time. Even in our intentionally simple model,our ndings are counter-intuitive: communicating too much informationabout the state of congestion can be detrimental to future congestion.Moreover, the population dynamics resulting from repeated interactions,where some information is withheld, while ensuring that the signals remainconsistent with the past observations, converge in distribution. 1 Introduction Congestion arises when many agents or users compete for limited resources,negatively a ecting each other’s use. Often, the performance of a particularresource for a particular user is inversely proportional to the number of con-current users. This occurs in many situations beyond the roads at rush hour,including power systems, ber optics in telecoms, and data centers. Notice thatcongestion is not only due to the inherent capacity limits of resources, but alsodue to agents choosing resources in a \synchronized" manner due to their lackof foresight into other agents’ choices.In this paper, we hence aim to \de-synchronize" agents’ actions by providingthem with signals. We model a repeated congestion problem as a multiagentsystem evolving over time, where the agents follow natural policies. We analysehow varying the amount of uncertainty agents face can reduce congestion andtune parameters of di erent signaling schemes accordingly. Further, we showthat the distribution of users across the resources is asymptotically stable, undera novel signaling scheme and mild further assumptions.In our model, a central agent has information about the state of the systemat all time instants, e.g., through sensor measurements, and can send signalsto other agents. We restrict ourselves to truthful signaling schemes, which are1
[1]
Toshihide Ibaraki,et al.
Resource allocation problems - algorithmic approaches
,
1988,
MIT Press series in the foundations of computing.
[2]
Y. She.
Urban Transportation Networks: Equilibrium Analysis with Mathematical Programming Methods
,
1985
.
[3]
R. Rosenthal.
A class of games possessing pure-strategy Nash equilibria
,
1973
.
[4]
Maria-Florina Balcan,et al.
The Price of Uncertainty
,
2009,
TEAC.
[5]
J. Harsanyi.
Games with randomly disturbed payoffs: A new rationale for mixed-strategy equilibrium points
,
1973
.
[6]
Tim Roughgarden,et al.
How bad is selfish routing?
,
2000,
Proceedings 41st Annual Symposium on Foundations of Computer Science.
[7]
R. Selten.
Reexamination of the perfectness concept for equilibrium points in extensive games
,
1975,
Classics in Game Theory.
[8]
J. Sobel,et al.
STRATEGIC INFORMATION TRANSMISSION
,
1982
.
[9]
James Hannan,et al.
4. APPROXIMATION TO RAYES RISK IN REPEATED PLAY
,
1958
.
[10]
Tim Roughgarden,et al.
The price of stability for network design with fair cost allocation
,
2004,
45th Annual IEEE Symposium on Foundations of Computer Science.
[11]
Michael Patriksson,et al.
A survey on the continuous nonlinear resource allocation problem
,
2008,
Eur. J. Oper. Res..
[12]
S. Stefanov.
Separable Programming - Theory and Methods
,
2001
.