Research On Data Center Network Traffic Forecasting And Load Balancing Based On Deep Learning

In recent years,with the continuous expansion of Internet technology and the widespread application of cloud computing technology,data centers have played an increasingly important role.As the core infrastructure of cloud computing technology,the data center is the center of data transmission,calculation and storage.The data center runs multiple types of services at the same time.Different types of services have different network requirements.Some services require high throughput,and some services are very delay-sensitive.A large amount of sudden traffic and rapid data flow forwarding make some links in the data center network have low utilization rates,but some links are frequently congested.When congestion occurs,the effective throughput of the network will decrease,which will affect the service performance and service quality of the data center.At the same time,the extensive use of virtualization also makes the traffic in the data center extremely complicated and difficult to control.The many-to-one transmission and communication mode causes TCP Incast problems,resulting in a sharp decline or even collapse of network throughput,which seriously affects the service performance of the data center network.Accurately predicting network traffic can reduce the frequency of network congestion,avoid network collapse,and ensure network fluency.By analyzing and predicting network traffic,you can grasp the characteristics and changing trends of network traffic in advance,and design corresponding traffic scheduling strategies based on the predicted results to avoid the occurrence of some network congestion and improve network performance to achieve the purpose of load balancing.In response to the above problems,this paper proposes a network traffic prediction model based on Particle Swarm Optimization(PSO)improved Long Short-Term Memory(LSTM)neural network,and a weighted balanced polling Weighted Balancing Round-Robin(WBRR)load balancing algorithm,specifically:1.In view of the problem that the parameters of the LSTM neural network model are usually difficult to determine,an IPSO?LSTM network traffic prediction model is proposed.The improved PSO algorithm is used to optimize the initial parameters of the LSTM neural network,and then the trained IPSO?LSTM model is used to conduct a comparative simulation experiment.The experimental results show that the PSO?LSTM model converges faster and improves the prediction accuracy.2.In view of the problem of poor performance of burst traffic load balancing in the data center network of the equal-cost multi-routing algorithm,the WBRR algorithm is proposed,which can effectively solve the link bottleneck and load fluctuations in the general path selection mechanism.Problems,thereby improving the bandwidth utilization of the network,increasing the throughput of the network,and reducing the delay of waiting in the queue.