Genetic algorithms execution control under a global application state monitoring infrastructure

Abstract

In the paper a new approach to the design of parallel genetic algorithms for execution in distributed systems with multicore processors is presented. The use of a distributed genetic algorithm based on new control implementation principles is proposed for an optimized irregular computational mesh partitioning for the FDTD (Finite-Difference Time-Domain) problem. The algorithm defines computational mesh partitions based on two objectives: load balancing and "min-cut" - the minimal number of edges between partition elements. The control in the parallel genetic algorithm assumes the use of program execution global control functions based on global application states monitoring. A control design infrastructure is provided to a programmer based on generalized synchronization/control processes called synchronizers. They collect local states of program computational elements, compute global control predicates and send back control signals. The paper describes how the assumed infrastructure can be used for convenient global program execution control at thread and process levels applied in the proposed genetic algorithm. © 2014 Springer-Verlag.