面向云原生的多租户混合调度模型研究

Research on Cloud-native Multi-tenant Hybrid Scheduling Model

云原生基础架构支持微服务、大数据、人工智能及高性能计算等多种负载类型,如何编排和调度多负载类型任务成为云原生技术应用面临的重要挑战。基于此,围绕云原生领域多负载类型的混合调度关键技术,分析面向云原生的多租户混合调度模型。首先,针对多租户场景下集群资源使用率低、各租户之间资源使用不均的问题,提出资源共享和资源独占模型;其次,针对混合调度资源分配不公平引起的服务质量问题,引入调度公平算法确保层次资源模型的调度公平性;最后,实现了基于Kubernetes的混合调度原型系统并验证了资源共享和资源独占模型的有效性和层次化模型的调度公平性。

The cloud native infrastructure supports multiple load types such as microservices,big data,artificial intelligence,and high-performance computing.How to arrange and schedule tasks of multiple load types has become an important challenge for cloud native technology applications.Based on this,analyze the key technology of hybrid scheduling with multiple load types in the cloud native domain,and analyze the multi tenant hybrid scheduling model for cloud native.Firstly,a resource sharing and exclusive model is proposed to address the issues of low cluster resource utilization and uneven resource utilization among tenants in multi tenant scenarios.Secondly,to address the quality of service issues caused by unfair allocation of mixed scheduling resources,a scheduling fairness algorithm is introduced to ensure the scheduling fairness of the hierarchical resource model.Finally,a hybrid scheduling prototype system based on Kubernetes was implemented and the effectiveness of the resource sharing and resource exclusive models,as well as the scheduling fairness of the hierarchical model,were verified.