Research On Workload Prediction Models In Cloud Computing Environments

Cloud computing industry has experienced a rapid development period since the birth of it.Now,more and more enterprises choose to deploy the companies' business in cloud computing centers and an increasing number of people choose to buy cloud computing services.Cloud service providers have got huge profits,but they are also under great pressure.The increase of business forced them to continue to expand cloud computing data center construction to meet the different needs of users.It also forced cloud Service providers to think about how to save cloud computing resources in the case of meeting users' resource requirements and not violating the SLA(Service Level Agreement).This is a great challenge for cloud service providers.There are a lot of discussions about the above problems in recent years.Some effective solutions have been proposed.For instance,researchers proposed virtual machine optimization strategies,capacity management strategies and user mode matching strategies,etc.These methods have helped cloud services providers improve the utilization of cloud computing resources and alleviate their pressure to some extent.In addition,the cloud computing workload forecast scheme has been proposed as another research direction.Researchers hope to predict the incoming cloud workloads in a future time period through studying the characteristics of workloads on cloud platform and combining the historical workload data.In this way,managers can configurate cloud computing resources reasonably according to the forecast results in time.So that cloud service providers can meet users' resource requirements and reduce the waste of resources at the sane time.Some researchers have acquired some achievement in this research orientation.But,there are some obvious disadvantages with existing schemes.Most prediction methods have low prediction accuracy rate.This will bring serious consequences to cloud users and cloud service providers.Inaccurate prediction results may lead too higher or too lower service configuration resource than actual usage resources.Too lower resources will result in tasks execution failure and servers crash down,and too higher resources will result in excessive waste of cloud computing resources.To summarize,existing prediction schemes are not enough mature so they are unsuccessful to meet cloud computing companies' demand.So that we should continue to work hard in the cloud workload prediction area.This paper proposes a novel cloud workload forecasting model.In this prediction model,we firstly use clustering methods to cluster historical workload data based on the analysis of characteristics of user behaviors.Similar load tasks will be put together through clustering algorithm.Followed by that,the processed data will be used to predict the future workload in a period of future time.So that manages can achieve the goal of the reasonable configuration of cloud computing resources with this prediction model.In this paper,the main work and creations are as follows:(1)In order to improve the prediction accuracy of load forecasting models,this paper analysis the load characteristics and user behavior characteristics,and expounds the relationship between them,based on the related historical load data in Google's cloud computing data center.This work provides a theoretical basis for the following load prediction model which is based on the classification of workload.(2)In order to accelerate the whole prediction progress of the load forecasting model,this paper improved the K-means clustering algorithm.In order to achieve the goal of quick clustering of load data,data density and the Quick Sort algorithm were introduced to the improved K-means algorithm so that it can quickly determine the initial clustering center in the process of clustering.(3)In order to further improve the efficiency and accuracy of load forecasting models,this paper proposes RVLBPNN prediction algorithm based on the improved BP neural network algorithm through the research about the neural network algorithm.This algorithm speeds up the process of data processing,and improves the accuracy of prediction through vector optimization.(4)This paper proposes a novel load prediction model called K-RVLBPNN through combining the improved k-means algorithm and RVLBPNN algorithm in view of the disadvantages of existing prediction models.The model makes full use of the load characteristics of user behavior.Moreover,the prediction performance of the novel cloud workload prediction model has been verified through comparing with HMM and NBC model.