Research On Fault Diagnosis Of Resource Pool Based On Data Analysis

Cloud Computing Resource Pool abstracts computer hardware into virtual resources and provides users with on-demandand scalable services.Reliability of Resource pools faces enormous challenges due to their inherent complexity and sharing.However most existing cloud platform fault diagnosis systems use thresholds for fault diagnosis,and there are deficiencies in adaptability and scalability.Therefore,it is necessary to research the fault diagnosis method of cloud platform resource pool.This thesis defines the behavior that the resource pool node deviates from the normal mode caused by unbalance load,performance bottleneck,and component anomaly as system fault.The cloud computing resource pool generates a large amount of node status data every day,and these data can characterize the state of the node.Thus this thesis mainly studies the method of fault diagnosis through data analysis.The main work of this thesis is as follows:(1)Study data collection method of cloud platform resource pool.This thesis obtains computing resource state data by injecting the console interface of the cloud computing operating system;based on the openflow protocol,an adaptive network traffic collection method based on multi-level flow table is designed,which measures the network bandwidth utilization,traffic,and packet loss rate by polling edge switch.(2)Different operating environments cause different node states,which ifluence the accuracy of fault detection.This thesis proposes a strategy of fault diagnosis based on environment division.Firstly Affinity Propagation clustering algorithm is used to divide the nodes of different operating environment.In order to adapt to the large scale mixed attribute dataset this thesis proposes a Fast IC_based Mixed attributes Affinity Propagation(FIHMAP),which divide the dataset into multiple subsets for clustering,and then combine the results to improve the efficiency of the algorithm.At the same time,the information cutting concept is introduced to maintain the accuracy of clustering.Secondly,separately model the divided subset of nodes to perform fault diagnosis.In order to be able to identify the known and unknown faults,an aggregated one-class support vector machine is proposed to perform multi-fault diagnosis.This method adopts the idea of semi-supervised,and multiple kinds of support vector machines are aggregated to distinguish a variety of faults.(3)Study the diagnosis method of association failure.This thesis applies the improved eclat association rule to mining anomaly pattern of resource pool under cloud environment,and futher improve the adaptability of the algorithm.Combined with fault diagnosis,it is applied to the mining and forecasting of associated anomalies.According to the experimental results on real datasets,the effectiveness of the proposed method is demonstrated.In particular,the fault diagnosis method based on environment perception is superior to the fault diagnosis method without environment awareness.In addition,the improved AP clustering and eclat algorithm The performance is better than the original algorithm.Aggregated one-class support vector machine can recognize known and unknown anomalies to obtain good fault diagnosis results.