Research On Information Loss And Burst Load Prediction Methods Based On Cloud Platform

In recent years,large-scale cloud computing data centers with high security,scalability and convenience have developed rapidly.It can provide paid resource services for enterprises and individuals according to demand.However,the popularity of cloud computing has also increased the pressure on data it faces.During the service provision process,it is inevitable that information loss,overload and no-load caused by burst load changes will occur.The accumulation of these problems can lead to cloud task failures or cloud server crashes.Therefore,it is particularly important to predict the problem of information loss and burst load in the cloud platform in order to take corresponding measures in advance.This article focuses on the analysis of the relationship between information loss and burst load changes and cloud task attributes in the cloud platform.In view of the current situation that traditional methods are difficult to predict information loss and burst load,a series of research on information loss and burst load prediction methods based on cloud platform are carried out by introducing multiple algorithms such as Gaussian mixture model(GMM),random forest and long short-term memory network(LSTM)into the construction of prediction models.(1)In view of the uncertainty of whether information loss can occur in cloud tasks and the type of loss that occurs,this paper first analyzes the type of information loss and focuses on exploring the relevance of cloud task attributes to information loss problems,and summarizes four task attributes with the highest relevance.Using these four attributes as experimental data,an information loss prediction model is established based on Gaussian mixture model(GMM).The balance factor is innovatively introduced to optimize and improve the prediction model combined with multi-element polynomial fitting algorithm.Multiple experiments show that the method proposed in this paper has high accuracy and small time complexity for predicting whether cloud tasks will have information loss and its loss type.Finally,a simple and effective optimization scheduling method is proposed for the prediction results.(2)In view of the irregularity of burst load changes,from a single task to a single machine,and then to a machine cluster,multi-granularity methods are used to analyze,classify and predict burst load phenomena.In the classification process,two-step discriminant classification models are innovatively proposed by combining three-decision theory with random forest algorithm.The burst load state of machines is divided into calm state,intermediate state and active state by manual rough classification and two-step discriminant method based on random forest.Long short-term memory network(LSTM)is used to establish a prediction model to predict the future burst load state of a single machine.Compared with multiple different neural network algorithms through experiments,it shows that the method proposed in this paper has high accuracy and small resource demand ratio,which can accurately predict the burst load state of a single machine in a period of time in the future.Finally,according to the prediction results,machines with the same state are clustered into machine clusters for load analysis to verify the accuracy of prediction.Simple and effective resource allocation methods are proposed for different burst load models combined with mixed deployment scheduling method.