Swarm-based distributed job scheduling in next-generation grids

Abstract

The computational Grid paradigm is now commonly used to define and model the architecture of a distributed software and hardware environment for executing scientific and engineering applications over wide area networks. Resource management and load balanced job scheduling are a key concern when implementing new Grid middleware components to improve resource utilization. Our work focuses on an evolutionary approach based on swarm intelligence and precisely on the ant-colony based meta-heuristic, to map the solution capability of social insects to the above resource scheduling and balancing problem, achieving an acceptable near-optimal solution at a substantially reduced complexity. The Grid resource management framework, will be implemented as a multi-agent system where all the agents communicate each other through the network and cooperate according to ant-like local interactions so that load balancing and Grid makespan/flowtime optimization can be achieved as an emergent collective behaviour of the system. We showed, by presenting some simulation results, that the approach has the potential to become really appropriate for resource balanced scheduling in Grid environments. © 2007 Springer.