Priority-grouping method for parallel multi-scheduling in Grid

摘要

This article presents a method of enhancing the efficiency of Grid scheduling algorithms by employing a job grouping method based on priorities and also grouping of Grid machines based on their configuration before implementing a suitable scheduling algorithm within paired groups. The Priority method is employed to group jobs into four groups, while two different methods, SimilarTogether and EvenlyDistributed, are employed to group machines into four groups before implementing the MinMin Grid scheduling algorithm simultaneously. Implementing the scheduling algorithms simultaneously within paired groups (multi-scheduling) ensures a high degree of parallelism, increases throughput and improves the overall performance of scheduling algorithms. Two sets of controlled experiments were carried out on an HPC system. Analysis of results shows that the Priority Grouping method improved the scheduling efficiency by very large margins over the non-grouping method.