Economic Markets to Regulate Mobile-Agent Systems

A Project of the D'Agents Laboratory at Dartmouth College

Jonathan Bredin, David Kotz, Daniela Rus

The value of any network is dependent on both the number of users and the number of sites participating in the network. There is little motivation for systems to donate resources to arbitrary agents, however. We remedy the problem by imposing an economic market on mobile-agent systems where agents purchase resources from host sites and sell services to users and other agents. Host sites accumulate revenues to be distributed to users to be used to launch more agents.

There are three major benefits from using markets to control mobile-agent systems. Hosts have incentive to entertain arbitrary mobile agents; markets enforce both temporal and spatial load balancing; and agents' limited budgets provide additional fault tolerance.

The flow of electronic currency in a market-based mobile-agent system.

An exchange of currency, services, and resources at a host site.

Currently we are using a resource allocation policy that allocates one resource to model agents' aggregate consumption. Servers solicit bids from agents for access. The bids represent the rate in currency per time unit at which an agent will pay for access to the server until the agent's job completes. The service that an agent receives is proportional to its bid relative to the sum of all competing bids.

We derive optimal bidding policies and allow agents to submit linear functions to express their policies. The function states what rate the agent will pay given the sum of the competing bids. Whenever the demand for service changes, the server searches for an equilibrium allocation that satisfies all agents' bidding functions.

These bidding policies determine the amount agents will spend at every site they visit. The agents can use the policies to estimate future expenditure and route themselves through the network to complete their itineraries.

The third paper listed below shows that priority queues do not scale well as the number user priority levels increases. The amount of resources left over from agents in higher priority levels decreases dramatically as the number of priority levels increases.

Relevant Dartmouth papers:

Jonathan Bredin, David Kotz, and Daniela Rus.  Mobile-agent planning in a market-oriented environment.Technical Report PCS-TR99-345, Dept. of Computer Science, Dartmouth College, May 1999. Revision 1 of May 20, 1999.

Jonathan Bredin, David Kotz, and Daniela Rus. Economic markets as a means of open mobile-agent systems.  In Proceedings of the Workshop ``Mobile Agents in the Context of Competition and Cooperation (MAC3)'' at Autonomous Agents '99, pages 43-49, May 1999.

Jonathan Bredin, David Kotz, and Daniela Rus.  Market-based resource control for mobile agents.  In Proceedings of the Second International Conference on Autonomous Agents, pages 197-204. ACM Press, May 1998.

Related projects inside D'Agents:

Monitoring and Modeling Dynamics of Information Space

Task Scheduling in Mobile-agent Systems

Relevant projects outside D'Agents:

Decision Machine Group, University of Michigan

Internet Ecologies Area, Xerox PARC

Yoav Shoham, Stanford University

Tuomas Sandholm, Washington University

Craig Boutilier, University of Toronto

Tamer Basar, University of Illinois Urbana Champaign