Simulation Studies of Multi-Criteria Resource Management Strategies in Grid Environments with Dynamic Descriptions of Jobs and Resources
The accuracy of the results of any performance study depends largely on the quality of the workload model driving it. While many research efforts have focused on the modeling of workloads, a little has been done in the context of comprehensive evaluation of different scheduling strategies, especially in the form of hierarchical parallel computing structures. We present results of simulation studies of scheduling algorithms using GSSIM in which both synthetic and realistic job workloads were used. We compare example scheduling algorithms in the light of many evaluation criteria relevant for end-users, e.g. total waiting time, and resource providers, such as utilization or load balancing. Based on this analysis, a new multi-criteria resource management model has been formulated to reflect hierarchical architectures of many complex parallel computing systems, their dynamic nature, and stakeholders preferences. We have proposed and tested a new multi-criteria heuristic called MGM which outperfors well-known scheduling strategies on some evaluation criteria. Our simulation studies can be used as a reference repository for further comparative studies.