Research And Application Of Cluster System Status Monitoring And Health Assessment Technology

With the development of the information age,the data infiltrates into various fields.In order to better utilize the data value,Developers from all walks of life begin to study data mining technology.The cluster plays an important role in data mining because of its excellent storage and computing abilities.With the continuous expansion of cluster size and more and more cluster service supports,cluster failures occur more frequently.In order to reduce the impact of cluster failures,cluster administrators want to be able to know the operating status and resource utilization of the cluster in real time.In order to shorten the failure time of the cluster system and improve the operation and maintenance efficiencies of the cluster,it is necessary to study the health monitoring and health assessment methods of the cluster system,which can help cluster administrators overall perceive the operation status of the cluster.The main work of the thesis includes the following points.First of all,because cluster monitoring parameters are too complicated for real-time understanding of operational status and other issues,the influence factors of cluster performance are studied and a new cluster monitoring performance metrics system is put forward and the calculation methods of metrics are given.Parameter acquisitions and processing solutions are designed too.Then,according to the characteristics of cluster support service,the influence weight of each monitoring metrics on cluster health is analyzed,and a health assessment method of cluster nodes is proposed.And based on the time series analysis method,a set of prediction models for performance metrics is proposed,which can predicts the performance metrics values and assesses cluster nodes health according to perfomance metrics values predicted.Finally,the experiments are designed to verify the proposed methods under actual production simulation environment.The experimental results show that the cluster status monitoring technology designed in this thesis can monitor the cluster performance and occupancy of resources,and can effectively give alarm when the cluster node performance metrics are abnormal.The method proposed to assess the cluster node health can calculate comprehensive health value of the node according to the node performance metrics value.And the proposed metrics prediction models can effectively predict the change trends of multiple performance metrics of the node.The innovative work in this thesis includes is following.(1)The proposed cluster performance metrics system and calculation method in this paper can simplify the metrics system under the premise of meeting the monitoring requirements and the expression of performance metrics can be clearer to be understood.(2)The proposed health assessment method considers the health of cluster nodes based on the characteristics of cluster support services and the comprehensive consideration of each performance metrics which achieves a comprehensive consideration of the health of cluster nodes.(3)The prediction models of performance metrics are established based on time series analysis to predict the performance of cluster nodes,which provide effective technical means for regularly grasping the trends of cluster changes.