Abstract: To evaluate job schedule on computational grids, the model of grid workload is constructed. On different nodes, job runtimes are different. Between different nodes, the costs of job migrations are different. Thus, the pure runtime and the pure migration cost are defined, which are independent of the performances of grid resources. The distributions of parallel degree, pure runtime, and arrival interval are obtained from workload models of parallel computers. Besides, the distributions of submittal location, pure migration cost, estimate factor of pure runtime, and deadline are constructed. Application instances show various workloads can be obtained from the model of grid workload to support comprehensive evaluation of job scheduling.

Key words: Computational grids, Job schedule, Workload model

摘要： 为评估计算网格中的作业调度，建立了网格工作负荷模型。在不同的节点，作业的运行时间不同；在不同的节点之间，作业的迁移开销不同。定义了不依赖网格资源性能的纯运行时间和纯迁移开销。借鉴并行计算机的工作负荷模型，可得到并行度、纯运行时间和到达间隔的分布。构建了作业提交位置、纯迁移开销、纯运行时间估计因子、完成期限的分布。应用实例表明，由网格工作负荷模型可获得 各种工作负荷，支持对作业调度的全面评估。

关键词: 计算网格, 作业调度, 工作负荷模型