Research On Flow Scheduling Method In Programmable Data Plane

Due to the explosive growth of traffic from applications and services in the data center network,traditional routing solutions are difficult to customize for an optimal scheduling solution.Software Defined Network(SDN)allows users to provide only their own control plane,but the data plane of the device remains irreplaceable and remains under the control of the vendor.Therefore,to address the above issues,a traffic scheduling strategy in the programmable data plane is proposed to exploit the high level of data plane programmability that the Programming protocol-independent packet processors(P4)language has to achieve traffic scheduling optimization.The study involves two main points as follows.1)An Imitation Learning based traffic scheduling approach is proposed.This method collects fine-grained real-time traffic data from the data plane of the SDN using P4-based in-band network telemetry.In the control plane,it combines a generative adversarial imitation learning model with a Soft-Actor-Critic(SAC)model to preserve traffic experience for better scheduling of large flows.2)A flow classification method based on binarized convolutional neural networks is proposed.The binarized neural networks are deployed in the data plane for traffic classification of different service types to ensure that the end-user or service-aware applications remain within acceptable limits.Finally,simulation experiments are performed for the two methods proposed above.The experimental results are compared with Equal-cost multi-path routing(ECMP),Hedera and Deep Deterministic Policy Gradient(DDPG)based traffic scheduling strategies under the same scenario of network environment,and analyzed by evaluating network performance metrics.Meanwhile,the experiments analyze the flow classification method based on accuracy,recall and F1 score evaluation criteria.The results show that the proposed method controls the link bandwidth utilization in the range of 10% to 80%,which guarantees the quality of service during network traffic transmission.Figure 18;Table 3;Reference 76...