Formalizing Distributed Search in Cooperative Distributed Problem-Solving Systems

University Of Massachusetts Amherst, Amherst MA

Investigators

Abstract

The explosive growth of computer networks has created the potential to access a great deal of computing power and information, but new models of distributed computing are required to fully utilize these resources. Cooperative distributed problem solving (CDPS) is concerned with solving large-scale problems using systems of intelligent software agents, distributed among networked computers, working cooperatively. CDPS approaches can produce highly flexible and reliable systems, and can be used for a wide range of potential applications. The goal of this project is to make it easier to to build CDPS systems for real-world applications, by improving the understanding of how system and domain characteristics interact to affect the performance of CDPS systems, and by identifying CDPS strategies that perform well under particular conditions. This will allow CDPS techniques to be more widely applied than they currently are. The project will pursue this goal by: developing formalisms for modeling CDPS systems as distributed search processes; implementing software tools that use these models to predict the performance of CDPS systems; identifying generic domain characteristics that are critical for predicting CDPS performance; and performing a broad range of experiments to build up a suite of effective abstract CDPS strategies.