Research On Resource Management Of Cloud-based Application System In Evolutionary Scenarios

With the development and rise of cloud computing,it is becoming more and more popular with users.As the virtualized product on the cloud,compared with traditional servers,cloud servers are characterized by high reliability,on-demand payment,elastic access to resources,and flexible management,so more and more web application managers choose to deploy their web application on the cloud.For administrators of a web application,on the one hand,they need to ensure the quality of service(QoS);on the other hand,they need to reduce the rental costs of cloud resources as much as possible while improving cost-effectiveness.Web application operators need to configure related computing resources for the cloud servers according to demand,where bandwidth resource is a major expenditure item in cloud service costs,and bandwidth is a relatively frequently updated resource.In order to ensure QoS,the operators need to predict the bandwidth demand in time according to the workload change and make the bandwidth adjustment schemes,and then evaluate the candidate schemes in advance to ensure that the selected scheme can provide stable and available services.Therefore,it is necessary to study bandwidth resource management.A candidate scheme can be regarded as a hypothetical system evolution.The research content of this paper is to conduct bandwidth driven what-if analysis for the cloud system.In existing studies of cloud-based application resource management,few studies focus on bandwidth resources,most of which focus on resources such as CPU and memory.Besides,most of the methods are not suitable for the prediction tasks in evolutionary scenarios and require adjusting datasets or retraining models.To solve these problems,we first present a what-if analysis method based on network simulation,which can be used to predict bandwidth demand and evaluate bandwidth management schemes.However,this method has many model parameters and complex adjustment process,so we then put forward a method based on machine learning,which is divided into two parts,one is to use traditional machine learning models for bandwidth demand and QoS prediction,the other is to use transfer learning method for feature transfer to help traditional machine learning models solve the problem of QoS prediction in evolutionary scenarios.Both methods proposed in this paper can help web application operators better manage bandwidth resources.The research contents of this paper are as follows:1)A network simulation-based method for predicting bandwidth resource requirements and QoS of web applications.This method uses a simplified parallel workload model,extracts parameters using automated log mining,and simulates complex network transmission processes using the network simulation tool to predict bandwidth requirements and changes in QoS under different workload intensities.Finally,we use TPC-W to validate the effectiveness of the method and evaluate several bandwidth management schemes.2)Because the stability of QoS prediction based on network simulation is slightly poor,and the model parameters are many and the adjustment process is complex,we propose a machine learning based method to predict network throughput and QoS.This method uses the traditional machine learning models,describes the data mining process in detail,and introduces bandwidth utilization as a feature to make the model get rid of the constraint of fixed bandwidth configuration.Moreover,the method also considers the impact of CPU resources.Finally,we also choose a cloud-based application based on TPC-W as a use case to evaluate the effectiveness and stability of the method and try to perform QoS prediction tasks in evolutionary scenarios.3)Because traditional machine learning methods cannot achieve good results for QoS prediction tasks in evolutionary scenarios,we then present a domain adaptation method for feature transfer to help traditional machine learning methods better predict QoS for web applications in bandwidth-driven evolutionary scenarios.This method classifies response time based on user experience and transforms regression problems into classification problems.Finally,we evaluate this method through comprehensive experiments for QoS prediction,including a variety of evolutionary scenarios under the benchmark system TPC-W and a real-world web application.The methods presented in this paper can be used to evaluate bandwidth management schemes for cloud-based web applications.The results of the evaluation can provide decision-making reference for web application operators,and help them provide high-quality application services while saving costs.Besides,the methods can also be applied to the management of other computing resources.