基于模拟退火算法的粒子群优化算法在容器调度中的应用

随着互联网产业的发展,虚拟机创建速度慢、不易扩展、灵活性不足等缺点越来越凸显,容器技术的出现为这些问题提出了一种新的解决思路;而现有的调度算法仅考虑容器云集群中工作节点的内存、CPU等物理资源,没有考虑对容器云调度后的镜像分发过程有明显影响的网络负载率,导致容器调度任务等待时间过长,造成数据中心的资源浪费;鉴于粒子群优化算法在局部开采能力和全局探测方面有较强的优势,提出了一种基于模拟退火算法的粒子群优化算法(SA-PSO,simulated annealing particle swarm optimization algorithm)的容器调度算法,通过使用模拟退火优化粒子群算法使其在算法初期跳出局部最优情况,提升算法性能;在Kubernetes平台实验过程中,SA-PSO调度算法相比Kubernetes的BalancedQosPriority算法,提升了整体节点资源利用率,显著减少任务最少等待时间;同时与标准PSO算法以及动态惯性权重PSO算法进行对比,不仅收敛能力有显著提升,并且相较标准PSO算法全局最优节点命中率提升近60％.

Application of Particle Swarm Optimization Algorithm Based on Simulated Annealing Algorithm in Container Scheduling

With the development of the Internet industry, shortcomings such as slow creation of virtual machines, difficult expansion, and insufficient flexibility have become more and more prominent. The emergence of container technology has proposed a new solution to these problems. The existing scheduling algorithm only considers the memory, CPU and other physical resources of the working nodes in the container cloud cluster, and does not consider the network load rate that has a significant impact on the image distribution process after the container cloud scheduling, resulting in too long waiting time for container scheduling tasks. Cause a waste of resources in the data center. Considering that the particle swarm optimization algorithm has strong advantages in local mining capabilities and global detection, a simulated annealing algorithm-based particle swarm optimization algorithm (Simulated annealing particle swarm optimization algorithm, SA-PSO) container scheduling algorithm is proposed. Using simulated annealing to optimize the particle swarm algorithm makes it jump out of the local optimal situation in the early stage of the algorithm, and improves the performance of the algorithm. In the process of the Kubernetes platform experiment, the SA-PSO scheduling algorithm compared to Kubernetes" BalancedQosPriority algorithm improves the overall node resource utilization and significantly reduces the minimum waiting time of tasks; at the same time, it is compared with the standard PSO algorithm and the dynamic inertia weight PSO algorithm, which not only converges The ability has been significantly improved, and compared with the standard PSO algorithm, the global optimal node hit rate has increased by nearly 60%.