工作量
计算机科学
分布式计算
公平份额计划
调度(生产过程)
排队
动态优先级调度
两级调度
单调速率调度
循环调度
作业调度程序
流水车间调度
实时计算
计算机网络
操作系统
数学优化
服务质量
数学
作者
Xiaofei Wu,Shoubin Dong,Liyun Zuo,Yizhen Sun
标识
DOI:10.1109/ispa/iucc.2017.00127
摘要
Mining job scheduling features based on extraction and analysis of workload trace in high performance computing clusters can be used to optimize scheduling strategy and enhance system performance. Based on detailed analysis of workload trace from a gene sequencing high performance computing system, this paper proposes a multi-queue backfilling scheduling algorithm, which is based on traditional backfilling scheduling. While optimizing for memory resource demands, this algorithm provides queue level load balancing to deal with the innate load imbalance characteristics of high performance systems. Experimental results based on practical gene sequencing workload trace clearly demonstrate that compared with traditional scheduling algorithms, the algorithm proposed in this paper is a good strategy to reduce the job waiting time and improve resource utilization.
科研通智能强力驱动
Strongly Powered by AbleSci AI