Abstract
Recently, numerous job scheduling techniques have been proposed for improving the efficiency of resource sharing in heterogeneous computational grids. However, as we have noticed, some scheduling algorithms, especially those working in batch mode, are likely to be suffered from deficiencies when extended to an non-dedicated grid environment, where the consistency degree of underlying computing environment is unpredictable. In this paper, we will present an ACO based load balancing approach to ameliorate the resource allocation in such a grid environment. Our load model derives from some typical non-dedicated computational grids (e.g. [9][13]) and we not only consider the heterogeneity of grid resources, but also take into account the overhead of job transferring among computing nodes. Compared with other load balancing algorithms (e.g. [6]-[8]) in the state of the art, our approach differs in the utilization of job granularity and the way load balancing is carried out. In addition, we have evaluated our approach by conducting simulations. The experimental results show that the proposed approach is efficient in promoting the system performance in terms of makespan and resource usage. The paper work is also conducted in the context of GORBA [5], a global optimising grid resource broker which is developed by our institute.