| Files are the basic form of data storage.With the advancement of enterprise informatization,data has become an important asset of current enterprises.In this context,building a distributed file storage system to achieve high-availability storage,management and access of massive files has become an industry hotspot.The purpose of this subject is to design and implement a highavailability distributed file storage system.Based on distributed nodes,the system realizes the functions of file multi-copy storage,elastic expansion of storage capacity,and elastic expansion of service capacity.In order to improve the availability of system data,the system adopts dynamic replica creation technology and dynamic replica positioning technology;in order to improve the availability of system services,the system adopts microservice architecture and trace management technology oriented to microservice architecture.In call chain management,the system introduces call chain-based exception detection technology to simplify system exception handling and improve operation and maintenance efficiency.This paper proposes an anomaly detection algorithm based on improved graph attention networks(GATs)for deviation prediction(improved GAT Deviation Prediction,iGAT-DP).This algorithm considers the call influence and resource influence relationship between different nodes in the microservice cluster.The algorithm first considers the calling characteristics and resource characteristics of microservice nodes,and obtains the mutual influence relationship between different microservices through the above characteristics to construct a weighted directed graph of influence relationship.Then,the node before the trace is used to predict the state of the next calling node by means of a sliding window,and the weighted directed graph of the influence relationship and the feature extractor are used in the prediction process to express the influence of other nodes in the microservice cluster on the calling node.Finally,it is judged whether the link is abnormal according to the deviation between the state of the predicted node and the state of the actual node.In the process of anomaly detection,the algorithm considers the interaction between different microservice nodes in the same microservice cluster,and realizes a more accurate judgment of link anomalies.A series of experiments show that the algorithm performs well,and its anomaly detection accuracy is improved by 7%to 10%compared to the baseline method.In order to verify the effectiveness of the system,the function and performance of the system designed in this paper are tested.The experimental results show that the system has high reliability and excellent performance.This paper first introduces the research background of the highavailability distributed file storage system;then,based on the investigation of various distributed file storage systems in the industry,analyzes the requirements of the high-availability distributed file storage system;The experimental verification is carried out;then the overall architecture design and each module design of the high-availability distributed file storage system are introduced,and the system implementation is introduced in detail;finally,the effectiveness of the system is verified through a series of tests. |
