| In recent years,with the rapid development of cloud computing technology,cloud-native technology has risen swiftly.Container orchestration systems,represented by Kubernetes,have been extensively adopted in cloud computing environments,becoming one of the core components of cloud infrastructure.However,with the increasing scale and complexity of cloud services,and recurring issues from cloud service providers,a single cloud computing service provider can no longer satisfy the tenants’ requirements for high availability of cloud services.Therefore,the adoption of multi-cloud architecture has become a significant trend in cloud-native technology.Multi-cloud deployments significantly enhance the availability of cloud services but inevitably increase the cost of multi-cloud networks.Moreover,due to the difficulty for tenants in effectively monitoring traffic variations in multi-cloud environments-a consequence of differences across multiple clouds-network costs have further increased,underscoring the urgency to address multicloud network cost issues.To efficiently optimize multi-cloud network costs,this paper proposes and validates a framework for efficiently collecting network packet information in kernel-mode from the operating system within Kubernetes,using the Extended Berkeley Packet Filter(e BPF)technology.After extracting features from the collected packet information,a mathematical model based on the Quadratic Assignment Problem(QAP)from combinatorial optimization is used.A multi-cloud network cost optimization model is proposed,solved by a combination of several heuristic search algorithms and random optimization algorithms.Although the QAP is proven to be an NP-hard problem and cannot yield precise results within polynomial time,the composite algorithm,integrating multiple optimization algorithm advantages,can find good approximate optimal solutions within a limited timeframe.To apply the near-optimal solution of multi-cloud network cost optimization to the workload,this paper proposes a multi-objective scheduling algorithm for multi-cloud computing tasks based on Deep Q-Network(DQN).This is designed to solve the problems of high latency in batch scheduling of computing tasks and imbalanced task scheduling in practical applications of Kubernetes.First,the algorithm incorporates the DQN state set input of prior states,enabling the system to fully leverage past experiences to guide future decisions.Next,the algorithm borrows the principles of immunity to optimize the reward function,enabling the algorithm to better understand and respond to changes in the environment,thereby choosing the optimal action policy.Lastly,the algorithm forms a balanced scheduling policy for computation tasks based on the idea of operator mutual exclusion,ensuring that computation tasks are distributed evenly among computing resources in a multi-cloud environment.Through these three strategies,the algorithm efficiently schedules computation tasks in a multi-cloud environment,thereby maximizing the utilization of computing resources.Upon experimental validation,the multi-cloud network cost optimization model and multiobjective scheduling algorithm for multi-cloud computing tasks based on DQN proposed in this paper have demonstrated outstanding performance.The model effectively reduces network resource costs in a multi-cloud,multi-region deployment environment within a controllable range of algorithm computation cost,and its performance in network cost optimization significantly surpasses that of the computing-resource-based scheduling strategy in the native Kubernetes scheduler.The research results of this paper have substantial practical application value,offering an effective method for multi-cloud tenants to optimize multi-cloud network costs. |
