一种改进的 DRF 算法对 BESIII 集群资源管理的优化

摘要

This paper presents an improved DRF resource allocation algorithm for BESIII cluster , as a result of the tests and research on different type of high energy physics jobs running on different machines .The machine performance rating and job type matching are added to the new algorithm as two factors for resource allocation . The new algorithm is implemented and tested with real data .The results show that the new algorithm can allo-cate resources more effectively , which improves the resource utilization and reduces the run time of jobs .%根据BESIII集群的实际需求，在测试和研究了不同类型的高能物理数据处理作业不同配置机器上的运行效果的基础上，提出了一种改进的DRF资源分配算法，加入了机器性能评级和作业类型匹配两个因素作为资源调度的依据，对算法进行实现并使用真实数据进行了实验测试。测试结果表明：新算法能够更加合理地分配资源，有效提高系统资源利用率，缩短作业运行时间。