基于动态限速的云计算应用负载调度方法-《计算机技术与发展》

文章信息/Info

Title:: Workloads Scheduling Approach for Cloud Computing Applications Based on Dynamic Rate Limit

作者:: 刘永波¹ ; 周摇博¹ ; 李亚琼 ¹ ; 李守超¹ ; 宋云奎²; 1. 江苏润和软件股份有限公司,江苏南京 210012;2. 中国科学院软件研究所,北京 100190

Author(s):: LIU Yong-bo¹ ; ZHOU Bo¹ ; LI Ya-qiong¹ ; LI Shou-chao¹ ; SONG Yun-kui²; 1. Jiangsu Hoperun Software Company,Nanjing 210012,China;2. Institute of Software,Chinese Academy of Sciences,Beijing 100190,China

Keywords:: dynamic rate limit; scheduling workloads; cloud computing applications; performance guarantee; resource management

摘要:: 云计算环境下,应用负载规模巨大,云服务提供商为多个客户提供共享的计算、网络和存储资源以最大化资源利用率,降低总体能耗,从而减少数据中心的运营成本,同时需要为用户提供良好的性能保障。针对该问题,提出一种基于动态限速的云应用负载调度方法。面向长时间运行的云应用,根据负载历史记录生成 r-b 曲线以描述存储和网络利用率,基于动态规划为每类负载自动生成限速参数。在保障处理负载的性能满足 SLO 的约束下,通过对自动化设置存储和网络限速参数,调度并整合负载以最小化处理负载的服务器数量,从而提高资源利用率并减少能耗。最后,设计并实现了原型系统,实验结果表明,提出的方法能够保障云应用性能,减少运行服务器数量,并具有较好的可扩展性。

Abstract:: In cloud computing environment,applications face a large scale of workloads. Cloud service providers provide shared computing,network,and storage resources to multiple customers to maximize resource utilization and reduce overall energy consumption,thereby reducing the operating costs of data centers and providing users with well performance assurance. The address this issue,we propose a dynamic rate limit-based workloads scheduling approach for cloud applications. For long-running cloud applications,r-b curves are generated based on load history to describe storage and network utilization,and speed limit parameters are automatically generated for each type of load based on dynamic programming. Under the constraint of SLO (service level object) to ensure the performance of the processing load,by setting storage and network speed limit parameters for automation,scheduling and integrating the load,the number of servers to process the load is minimized,thus improving resource utilization and reducing energy consumption. Finally,we have designed and implemented a prototype system. The experiment demonstrates that the approach proposed can guarantee the performance of cloud applications and reduce the number of running servers with well scalability.