Research On Traffic Scheduling Method Based On Machine Learning

With the development of network technology,network applications continue to increase,and various services have an increasing demand for bandwidth.In order to meet the transmission requirements of various services,it is necessary to continuously increase the network bandwidth.Since the total network bandwidth is limited by physical equipment and there is an upper limit,it is impossible to increase the bandwidth indefinitely to meet the needs of all services.In view of the above-mentioned limited bandwidth resources,traffic scheduling can be performed on services so that the service data flows through the network are transmitted according to specified rules or sequences,so that the utilization of network resources can be improved while ensuring the normal communication of various services.Traffic scheduling is mainly responsible for allocating corresponding network resources for each business according to business needs.By analyzing the transmission of business data flows in the network,the corresponding queue scheduling scheme is implemented,so as to meet the business transmission needs as much as possible.Provide services for more businesses in the case of network resources.Based on the above situation,this paper proposes a traffic scheduling method based on machine learning.The main tasks are: 1)By studying the existing traffic prediction methods based on machine learning algorithms,a network traffic prediction model based on random forest is proposed.The model extracts the three cycle characteristics of the data stream's cycle starting point,cycle size,and cycle similarity by extracting the statistical characteristics of the network data flow,and predicts the data value of each time sequence point according to the location feature of the time sequence point,that is,the predicted flow rate Value;2)Combining the prediction results of the network traffic prediction model and the weighted balance round-robin queue scheduling algorithm,a cycle-based queue scheduling algorithm is proposed.The algorithm is based on each timing point of different data flows at the beginning of each cycle.According to the predicted traffic value,the maximum and minimum fair allocation principle is adopted to allocate corresponding bandwidth to the queues where different data flows are located,and the weight value of queue scheduling is designed according to the bandwidth value.When the time enters the next cycle,the bandwidth value of the data stream is updated,and the weight of each data stream is adjusted according to the updated bandwidth demand,so as to perform different dequeue operations for different queues.Finally,this article uses the above method to design the corresponding traffic bandwidth determination module and queue scheduling module on the business service system platform developed in the laboratory,and realizes the traffic prediction function and the traffic scheduling function,and conducts experimental tests on the two modules.The analysis of the simulation results verifies the feasibility of the scheme.