Research On Optimization Method Of Elastic CPU Resource Management For Container Environment

The container is a lightweight virtualization technology that has been widely used in cloud computing.Compared with the traditional virtual machine,it has faster startup and deployment speed,lower operating overhead,and higher deployment density.Elastic resource management is an essential feature of cloud computing platforms.It can dynamically scale resource configurations to respond to load changes without interrupting services,thereby improving resource utilization while ensuring the quality of service for applications.However,the existing elastic CPU resource management mechanisms in the container environment are inadequate.First,applications in the container are unaware of the elastic changes of the container CPU resource.They cannot fully use the elastic CPU resource,resulting in a decrease in CPU resource utilization.Second,the container CPU resource elastic management mechanisms lack consideration of CPU architecture characteristics(e.g.,Non-Uniform Memory Access,NUMA),resulting in severe performance degradation.Research on the elastic CPU resource management mechanism for container environment is carried out,and optimization strategies are proposed for the above two issues.Firstly,based on a large number of experiments,an application parallelism setting strategy based on the container CPU resource elasticity is proposed.The strategy can set appropriate parallelism for the application and optimize the thread blocking synchronization mechanism so that the application can adapt to the elastic changes of the container CPU resources,thus improving the application performance.Secondly,a NUMA-aware container CPU resource scheduling policy is proposed.It can allocate appropriate CPU resources to containers based on NUMA topology and adjust resources in real-time according to load changes,effectively reducing remote memory access and improving container performance.Finally,a containeroriented elastic CPU resource management system is designed and implemented to integrate the above two optimization strategies,thus improving the container performance effectively.Experimental results show that the application parallelism setting strategy based on the container CPU resource elasticity can shorten the application execution time by up to 52%and 18% under low and high system load scenarios,respectively.Compared with the popular NUMA-aware scheduling policy,the NUMA-aware container CPU resource scheduling policy can shorten the application execution time by up to 28% and 27% for the individual container and the related container group.Compared with the popular scheduling strategy,the container-oriented elastic CPU resource management system,which integrates the above two strategies,can reduce application execution time by up to 47% and 30% under low and high system load scenarios,respectively.