应用感知的容器资源调度优化方法  

Application-Aware Container-Oriented Resource Scheduling Optimized Approach

作　　者：陆志刚[1,2,3] 吴悦文 顾泽宇[1,2] 吴启德[4] LU Zhi-Gang WU Yue-Wen GU Ze-Yu WU Qi-De(Institute of Software, Chinese Academy of Sciences, Beijing 100190, China University of Chinese Academy of Sciences, Beijing 100049, China SIP State Property Holding Co. Ltd., Suzhou 215028, China University of Science and Technology of China, Hefei 230026, China)

机构地区：[1]中国科学院软件研究所,北京100190 [2]中国科学院大学,北京100049 [3]苏州工业园区国有资产控股发展有限公司,苏州215028 [4]中国科学技术大学计算机科学学院,合肥230026

出　　处：《计算机系统应用》2017年第3期134-138,共5页Computer Systems & Applications

摘　　要：资源调度作为容器管理的关键技术之一,已有研究工作或满足公平性目标,将工作负载平均调度到所有物理节点上,关注吞吐率指标;或满足性能目标,将工作负载关联的多个容器载体调度到相同或相近的物理节点上,关注响应时间.提出应用感知的容器资源调度方法,采用多队列模型兼顾公平性和性能目标.实验结果显示,对于典型的大数据处理场景,本方法和已有公平性调度方法具有相当的吞吐率;对于典型的事务型应用场景,本方法相对于已有的公平性调度方法,事务型应用的延迟最多可减少100%.Resource scheduling is a key technique for container management. Prior work or meets the goal of fairness, the work load average is scheduled to all physical nodes, pays attention to the throughput indicator; or to meets performance targets, multiple carrier scheduling the workload related to physical nodes in the same or similar, pay attention to the response time. In this paper, it presents an application-aware resource scheduling approach, and employs a multi-queue model to meet both fairness and performance targets. Experimental results show that the approach has equal throughput for typical big data processing scenarios, and can reduce latency by up to 100% for typical transactional application scenarios.

关 键 词：应用感知 容器 资源调度 大数据 事务型 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]