Research On Load Balancing Method Of Data Center Based On Time Series Prediction

With the rapid increase of data generated in the Internet,the distribution of requests for data storage and use of server cluster system has been increasingly common phenomenon,and then draw is the load balancing problem in cluster system,due to different users to access nodes in frequency,high frequency of nodes will lead to an infinite long request queue,waste of resources of low frequency node the main content of this article is to make a research on load imbalance problem:Firstly,this paper constructs a time series prediction model based on LSTM(Long Short-Term Memory)Neural network,and proposes a time series prediction model based on C-L(Convolutional Neural Network-Long Short-Term Memory)Neural network to solve the problem that the prediction accuracy of this model is insufficient in the face of sudden burst of network traffic In this model,the first to use a one-dimensional convolutional neural network to extract the time correlation characteristics,trained by the outbreak of the flow characteristics of the burst characteristics,after being mixed with data sets using LSTM forecasting model to forecast data use wikipedia clicks simulation experiments show that the data set based on the C-L time series prediction model of neural network in the face of the situation of the network traffic burst has good performance,the model prediction accuracy is higher than the traditional method of average and LSTM predictions of the model.Secondly,aiming at the limitations of the current typical algorithm polling method,weighted polling method and smooth weighted polling method in load balancing,this paper proposes a CLDC(continuous neural networks long short term memory dynamic)algorithm The load balancing algorithm introduces the load impact vector into the consideration of request allocation,and takes the current CPU utilization,disk I / O utilization,memory utilization and network bandwidth of the server as the load impact vector.The algorithm obtains the load impact vector of each server in the cycle,and sets the load weight for each server.According to the predicted concurrency of time series prediction model based on C-L neural network and the ideal concurrency quantity of server cluster,the cycle is dynamically adjusted to make the cycle change with the change of load.The simulation experiments using Wikipedia click data set show that the CLDC load balancing algorithm has better response time and network throughput than weighted polling method and load vector based polling method.