Research On Load Balancing Algorithm For Identification Resolution System Of IIoT

With the development of the identification resolution system of industrial internet of things（IIoT）,the amount of registration and resolution of identification will reach a massive level.High concurrency of identity service requests will also bring great challenges to the second-level node system.The load balancing algorithm can distribute the network traffic to each node evenly.However,based on some static data,a constant proportion is mainly used to allocate concurrent requests by the traditional static load balancing algorithm.Without considering the real-time running state of the server and the burst of the traffic,the purpose of load balancing cannot be achieved after the node has been running for a long time.Therefore,this dissertation first conducts related research on the neural network prediction model,and predicts the network traffic of the secondary node through the prediction model.Therefore,this dissertation investigates the research status and characteristics of load balancing algorithm,and studies the basics of related technologies,including long short term memory（LSTM）network and deep learning.Furthermore,the combination of LSTM algorithm and dynamic load balancing algorithm is deeply explored and studied.A dynamic load balancing algorithm of self-adaptive identification analysis node is proposed,which can sense the real-time running state of the server and traffic burst.The specific work and main innovation points are as follows:（1）A load balancing system for identity resolution nodes is designed.In the second-level node system of identity resolution,the second-level nodes of an industry are distributed in different areas.In order to spread the concurrent requests to multiple nodes,a load balancing system is designed and deployed in each second-level node.The main function modules of the system can be divided into data acquisition module,load balancing algorithm module,traffic scheduling module and so on Cluster management module.（2）A prediction model for the number of requested connections based on LSTM is proposed.First,the number of historical request connections of the cluster is selected as the data set,and the data set is standardized.Taking into account the time series characteristics of the data set,the LSTM neural network is selected as the traffic prediction model to predict the number of requested connections in the cluster at the next moment,and to map the network burst traffic of the nodes.（3）A weighted minimum connection scheduling algorithm based weight calculation method is proposed.The algorithm takes the weight of the node as the selection criterion of the node,and takes the future network traffic of the node,the performance index of the node,the real-time load status of the node,and the network traffic of the node at the current moment as the weight evaluation parameters of the node,and obtains through these parameters A method of calculating node weights.Simulation experiments show that the LSTM traffic prediction model has excellent performance for processing time series data,and has a more accurate prediction effect than traditional mathematical statistical models.Compared with traditional load balancing algorithms,the IIoT load balancing algorithm not only has a higher response success rate,but also reduces the average response time of concurrent requests processed by the cluster by about 50%,and increases the throughput rate of the cluster by about 30%on average.