【摘要】 云计算的不断发展，微服务部署对计算机硬件以及软件的交互过程提出了更高的要求，通过容器调度的方式来处理用户对应用服务提交的请求已经成为了行业范式。针对容器调度过程中存在的负载不均衡、服务硬件要求多元化等问题，本文提出一种基于竞争双深度神经网络强化学习算法的资源控制容器调度优化算法(RCCS-D3QN)。通过实时采集集群各类资源的消耗情况并结合权重因子，利用资源满足率和集群资源均衡率，构造奖惩函数，深度强化学习模型。利用仿真平台RayCloudSim构建智能体交互环境，对算法中的神经网络进行训练，基于实际微服务项目来验证结果的有效性。实验结果表明，RCCS-D3QN算法对比传统调度算法在保证服务质量的同时，在资源利用率、负载均衡度方面有显著提高。更多还原

【Abstract】 With the growing development of cloud computing, higher requirements have been put forward for the utilization of computer hardware and software, The deployment of microservices through container encapsulation has become an industry paradigm. This article focuses on load imbalance and diversified service hardware requirements in container scheduling, propose a resource control container scheduling optimization strategy RCCS-D3QN based on Dueling Double DQN reinforcement learning algorithm. By collecting the consumption of various resources in the cluster in real time and combining it with the weight factor, using the resource satisfaction rate and the cluster resource balance rate, a reward and punishment function is constructed to implement the reinforcement learning model. By building an interactive environment through the simulation platform RayCloudSim and training the algorithm network, which verify the effectiveness of the results based on actual microservices. The experimental results show that compared with the traditional scheduling algorithm, the RCCS-D3QN algorithm can significantly improve resource utilization and load balancing while ensuring quality of service.更多还原