Sub optimal scheduling in a GRID using genetic algorithms

The computing GRID infrastructure could benefit of techniques that can improve the overall throughput of the system. It is possible that job submission will include different ontology in resource requests due to the generality of the GRID infrastructure. Such flexible resource request could offer the opportunity to optimize several parameters, from network load to job costs in relation to due time, more generally the quality of services. We present the result of the simulation of GRID jobs allocation. The search strategy for this input case do not converge to the optimal case inside the limited number of trial performed, in contrast with previous work on up to 24 jobs. The benefits of the usage of the genetic algorithms to improve the quality of the scheduling is discussed. The simulation has been obtained using a software environment GGAS suitable to study the scheduling of jobs in a distributed group of parallel machines. The modular structure of GGAS permit us to expand its functionalities to include other first level schedule policies with respect to the FCFS that is considered. The result of this paper suggest the usage of a local search strategy to improve the convergence when the number of jobs to be considered is big as in real world operation.