Research Of File System Metadata Graph

With the rapid development of Internet technologies,the data show explosive growth.The metadata access becomes the bottleneck in the massive data storage systems.At present,most storage systems use the tree structure namespace to organize and manage metadata or use relational database to store metadata and improve the efficiency of metadata access through the index.However,these traditional metadata management methods ignore the correlation between files,which leads to a single metadata management function and can not provide richer and more personalized service.In this paper,based on the research of the distributed file system HDFS,we design and implement a metadata management method based on graph structure.Firstly,according to the characteristics of file data to extract the specific semantic information,using the semantic information to construct semantic metadata,then organizing the file metadata and semantic metadata into metadata graph according to the semantic correlations between the different files.And transformation of the metadata service node Namenode in HDFS,we design a new metadata service node Graphnode which provides services through metadata graph,finally,we use the collaborative filtering,node influence analysis and cluster analysis algorithms to mine the associations between the file data,so that Graphnode can provide richer and more intelligent services.The test results show that the implementation of Graphnode based on metadata graph can effectively manage the metadata and the relationships between metadata.While providing traditional metadata services,using the rich semantic information and the relationship among vertices in the metadata,Graphnode can perform data mining and analysis in the metadata layer.Compared with the traditional file system to achieve file recommendation and clustering analysis,the performance has been improved several times.This not only provides a solution to the intelligent storage system,but also makes the organization and management of metadata more meaningful and valueful,and provides a new solution to the data mining and analysis of big data era.