机群环境中基于信度模型的作业调度算法

Job scheduling algorithm based on credit model in cluster environment

提出了一种新的机群环境下的作业调度算法 ,该算法基于信度模型 .在信度模型中 ,每一个计算节点是一个信度实体 ,系统根据信度实体的物理能力、当前状态、历史作业完成情况 ,对节点进行信度评价 ,赋予其相应的信度值 ,调度算法根据信度实体信度值为作业分配合适的资源运行作业 .同时 ,调度系统还对机群系统的整体信度进行评价 ,并根据估计的作业运行时间和完成期限对作业做信度要求折算 ,调度系统根据机群系统的整体信度情况和作业信度要求来决定是否接受该作业请求 .在OpenPBS机群调度系统上实现了该算法 ,并通过GridSim工具包与先入先出FIFO(First InputFirst Output)算法和Libra调度算法进行了模拟对比试验 .试验结果表明 ,该算法是有效的 .

A novel job scheduling algorithm based on credit model was presented for cluster system. In the credit model, every computing node was looked as a credit entity. The scheduling system gave each entity a credit value according to its power ability, status and history records. The scheduling system also gave a credit estimate to the entire cluster, and according to this estimate and entering job request it decided to reject or accept the job. It was realized based on OpenPBS framework. It was also simulated using GridSim toolkits to proceed a contrastive test with FIFO (first-input first-output) and Libra. The result shows that it works well.