容器编排工具的出现使得容器虚拟化技术解决了云平台资源利用率低、调度分发缓慢等诸多问题，因此，研究容器编排工具的调度算法可以有效分配容器集群资源，提高资源利用率，同时最小化资源消费总成本。为得到更好的调度策略，论文使用元启发式算法对主流容器编排工具——Kubernetes的调度策略存在的问题展开研究。 针对Kubernetes调度模型在部署长时运行服务时未考虑资源消费成本和多Pod部署未考虑均衡调度的问题，提出改进的蚁群粒子群算法优化基本调度模型。首先建立成本和集群负载度的目标函数；然后通过蚁群粒子群算法的节点寻优策略改进基本调度模型的节点选择策略；最后在CloudSim平台上实现四种调度模型：蚁群、粒子群、蚁群优化粒子群和基本的调度模型。 针对以上四种调度模型在集群资源量不足时均存在调度失败率高的问题，提出一种新的抢占式调度策略。首先根据Pod的运行态、终止态和重启策略设计Pod的优先级；其次根据抢占式机制设计新的抢占式调度模型。 针对蚁群粒子群算法中节点寻优耗时的问题，提出基于Go调度模型原理实现的并行化算法来优化节点寻优效率。首先在IntelliJ IDEA平台上对Goroutine并行度进行测试，找出最优并行度；然后在最优并行度的基础上对蚁群、蚁群优化粒子群和粒子群优化蚁群算法的节点寻优与计算过程进行并行化设计；最后在优化的目标函数基础上，对基本调度模型和并行化调度模型进行实验对比。 实验表明：论文设计的蚁群、蚁群优化的粒子群调度模型相比基本调度模型不仅节约资源消费成本，而且降低多Pod调度时集群的负载度；抢占式调度策略在集群资源量不足时使得尽可能多Pod成功调度，降低Pod的调度失败率；在Goroutine最优并行度的实验环境中，并行蚁群算法相比串行蚁群算法的运行时间缩短了近2到3倍，且并行化蚁群调度模型相对其他调度模型分配效果最佳。

The emergence of container orchestration tools has enabled container virtualization technology to solve many problems such as low utilization of cloud platform resources and slow scheduling and distribution. Therefore, the scheduling algorithm of the container orchestration tool is studied, which can effectively allocate container cluster resources, improve resource utilization, and minimize the total cost of resource consumption. In order to get a better scheduling strategy, the paper uses meta-heuristic algorithm to solve the problems in scheduling strategy of the mainstream container orchestration tool-Kubernetes. For the Kubernetes scheduling model, when the long-term running service is deployed, the resource consumption cost is not considered and the multi-Pod scheduling does not consider the cluster load degree. The ant colony particle swarm optimization algorithm is proposed to optimize the basic scheduling model. First, establish the objective function of cost and cluster load;Then, the node optimization strategy of the basic scheduling model is improved by the node optimization strategy of the ant colony particle swarm algorithm;Finally, four scheduling models are implemented on the CloudSim platform: ant colony, particle swarm, ant colony optimization particle swarm and basic scheduling model. A new preemptive scheduling strategy is proposed for the above four scheduling models with high scheduling failure rate when the cluster resources are insufficient. Firstly, the priority of Pod is designed according to the running state, termination state and restart strategy of Pod. Secondly, a new preemptive scheduling model is designed according to the preemptive scheduling mechanism. Aiming at the problem of time-consuming in ant colony particle swarm algorithm, a parallelization algorithm based on Go scheduling model principle is proposed to optimize node optimization efficiency. Firstly, the Goroutine parallelism is tested on the IntelliJ IDEA platform to find the optimal parallelism; Then, based on the optimal parallelism, the ant colony, ant colony-optimized particle swarm and particle swarm-optimization ant colony algorithmare are designed to parallel the process of node optimization and calculation. Finally, based on the optimized objective function, the experimental comparison between the basic scheduling model and the parallel scheduling model is carried out. Experiments show that the ant colony and ant colony-optimized particle swarm scheduling model designed in this paper not only saves cost, but also reduces the load degree of the cluster when multi-Pod scheduling;The preemptive scheduling policy reduces the scheduling failure rate of the Pod when the cluster resources are insufficient.In the experimental environment of Goroutine optimal parallelism, the running time of the parallel ant colony algorithm is nearly 2 to 3 times shorter than that of the serial ant colony algorithm. And the parallel ant colony scheduling model has the best Scheduling effect compared with other scheduling models.