Dynamics of computational ecosystems

Abstract

Recently, Huberman and Hogg [in The Ecology of Computation, edited by B. A. Huberman (North-Holland, 1988), pp. 77–115] analyzed the dynamics of resource allocation in a model of computational ecosystems which incorporated many of the features endemic to large distributed processing systems, including distributed control, asynchrony, resource contention, and cooperation among agents and the concomitant problems of incomplete knowledge and delayed information. In this paper we supplement an analysis of several simple examples of computational ecosystems with computer simulations to gain insight into the effects of time delays, cooperation, multiple resources, inhomogeneity, etc. The simulations verify Huberman and Hogg’s prediction of persistent oscillations and chaos, and confirm the Ceccatto-Huberman [Proc. Natl. Acad. Sci. U.S.A. 86, 3443 (1989)] prediction of extremely long-lived metastable states in computational ecosystems. Extending the analysis to inhomogeneous systems, we show that they can be more stable than homogeneous systems because agents with different computational needs settle into different strategic niches, and that overly clever local decision-making algorithms can induce chaotic behavior.