针对大型数据中心复杂的软件定义网络(SDN)架构中多控制器结构引发的负载不均衡问题,提出了一种基于强化学习的交换机迁移策略。首先,综合考虑负载均衡度与控制器负载分布,将交换机迁移问题建模为组合优化问题。其次,结合基于SumTree的优先迁移机制对SAC算法进行优化,以最大化负载均衡度的改善,同时采取较小迁移开销的策略。借助Server建立全局控制器的控制平面连接,以根据负载状态实施交换机迁移,最终实现控制器的负载均衡。仿真结果表明,该策略有效地依据负载状态实现了负载均衡。在简单负载环境中负载均衡度提升了17.34%；在复杂负载环境中,性能提升更为显著,达到74.45%,同时在迁移开销方面也表现出一定优势。

In response to the issue of load imbalance caused by the multi-controller architecture in complex Software Defined Network (SDN) structures within large data centers, a switch migration strategy based on reinforcement learning is proposed. First, the switch migration problem is modeled as a combinatorial optimization problem, taking into account both load balancing and the distribution of controller loads. Next, we optimize the Soft Actor-Critic (SAC) algorithm by incorporating a priority migration mechanism based on a SumTree, aiming to maximize improvements in load balancing while employing a strategy that incurs minimal migration overhead. A global control plane connection is established through a server to facilitate switch migration based on load conditions, ultimately achieving load balancing among controllers. Simulation results indicate that this strategy effectively realizes load balance according to load states. In a simple load environment, load balancing improved by 17.34%; in a complex load environment, the performance enhancement was even more significant, reaching 74.45%, while also demonstrating certain advantages in terms of migration overhead.

TP393???????

国家自然科学基金项目（面上项目，重点项目，重大项目）,