Stability in referral systems

Issue title: Engineering Environments in Multiagent Systems

Guest editors: Danny Weyns and Andrea Omicini

Article type: Research Article

Authors: Gürsel, Anıl | Sen, Sandip^{; *} | Candale, Teddy

Affiliations: University of Tulsa, Oklahoma, USA

Correspondence: [*] Corresponding author. E-mai: sandip-sen@utulsa.edu

Abstract: We evaluate the effect of referral mechanisms on stability of agent populations using service providers. We consider a pool of service providers whose performance varies inversely with load. An agent's satisfaction with a service provider depends on a private function of the performance of the provider. An agent will ask for a referral from a centralized referral agency if its satisfaction from its current service provider is below its satisfaction threshold. Such an agent chooses one of the referred providers in the next time interval. The referral system reaches stability when all agents are satisfied with their current provider. We evaluate the relative merits of the following decisions by the centralized referral agency: number of service providers returned and the means for choosing those referrals. The performance metric used is the number of time periods to reach stable system configurations. Interesting observations include the advantage of performance proportionate and semi-deterministic schemes over deterministic referral decisions.

DOI: 10.3233/MGS-2009-0117

Journal: Multiagent and Grid Systems , vol. 5, no. 1, pp. 19-36, 2009