【摘要】 目前SRv6网络中的流量调度方法主要是基于固定或启发式规则的方法,缺乏灵活调度整体网络流量的能力,难以适应动态的网络环境变化。针对SRv6网络缺乏关键流识别能力的问题,文章提出一种基于深度强化学习的关键流识别算法,建立适应网络动态变化的关键流学习模型,在不同的流量矩阵中识别出对网络性能影响最大的关键流集合。针对SRv6网络流量调度问题,文章提出一种基于关键流的流量调度优化算法,采用线性规划求解出每一条关键流的最优显式路径,并采用不同的路由方式对普通流和关键流进行负载均衡。实验结果表明,该算法可显著提升SRv6网络流量负载均衡能力,降低网络端到端传输延迟。更多还原

【Abstract】 Current traffic scheduling methods in SRv6 network are mainly based on fixed or heuristic rules, which lack of the ability to schedule overall network traffic flexibly and are difficult to adapt to dynamic network environment changes. To address the deficiency in key flow identification within SRv6 network, the article introduced a key flow identification algorithm based on deep reinforcement learning. This approach established a key flow learning model adapted to the dynamic changes of the network, identifying sets of key flows that significantly impact network performance across various traffic matrices. In response to the challenges of traffic scheduling in SRv6 network, the article developed an optimization algorithm for traffic scheduling, rooted in key flow analysis. This algorithm employed linear programming to determine the optimal explicit path for each key flow and utilized different routing methods for ordinary flows and key flows, effectively enhancing network performance. The experimental results demonstrate that the proposed traffic scheduling algorithm leads to a significant improvement in network load balancing and a substantial reduction in network end-to-end transmission delay.更多还原