面向算力网络的多路径时敏优先调度机制

摘要：研究了一种面向算力网络的多路径低时延转发调度算法。该算法可以根据网络状态变化动态更新路径价值量生成路径转发决策，并在一定置信概率内以多路冗余发包的方式进行多路备份传输，降低路径传播时延。还提出了一种等级与队列映射算法，利用网络可编程技术改进设备转发逻辑，利用有限数量严格优先队列保障数据包近似按等级出队，降低数据排队时延。仿真实验结果表明，所提出方法可以降低数据传输时延及抖动，为算力网络业务提供稳定吞吐量。

关键词：算力网络；低时延；多路径；优先队列

Abstract: A multipath low latency forwarding scheduling algorithm for computing power networks is studied, which dynamically updates the path value to generate path forwarding decisions based on network state changes. Within a certain confidence probability, multiple backup transmissions are carried out in the form of multiple redundant contracts to reduce path propagation delay. In addition, a rank and queue mapping algorithm is proposed, which uses network programmable technology to improve the device forwarding logic and uses a limited number of strict priority queues to ensure that packets are approximately queued according to the rank, reducing the data queue delay. Simulation results show that the proposed method can reduce data transmission delay and jitter, and provide stable throughput for computing network services.

Keywords: computing power network; low latency; multipath; priority queue