An Experimental Study of Systemic Behavior of Cooperative Search Algorithms

Abstract

Parallel and distributed computer systems are increasingly used to find good solutions to difficult combinatorial optimization problems. An interesting and often used approach consists of executing concurrently different search methods that exchange information gathered in previously explored regions of the solution space. Cooperation among sequential programs strongly affects how the solution space is explored and it is hoped that it improves the efficiency of the search. A formal representation of these cooperative algorithms based on a discrete-time dynamical system model may be used to describe how the search behavior of cooperative programs depends on systems of complex and correlated interactions. We define systemic cooperation as the search behavior for programs interacting within the same system of correlated interactions, and show experimentally the importance of the phenomenon and its independence of specific search design parameters.