Research On Multi-attribute-based File Organization Mechanism

The rapid development of computer technology brings the emergence of variouskinds of data resources. It is difficult for user to analyze and layering the data structureand layer, so we are facing a serious problem witch is about how to organize fileseffectively. Nowadays, the mainstream of file systems are concerned about theperformance and reliability problems in data storage filed, while the connection betweencontent of the data itself and the semantic information was less valued. To solve theproblems above, the academic extended the semantic relationships between files byintroducing a new model in existing file system. However, the file system structure wasnot modified completely; query performance and portability were not considered seriously.Therefore, it’s quite meaningful to study and realize an efficient file management system.By analyzing and discussing the three types file organization mechanism and fileretrieval mechanism, we summarize the characteristicsof the management andshortcomings of current file systems. Then, we reorganize file’s structure by opening up aextended attribute region, so the self-description of file can be enhanced.At last, weimplement a prototype system based on above. The information is organized as a flat filestructure in the prototype system logically. We provide a mechanism that compatible withthe POSIX specification to achieve access by virtual path，we realize some flexibleinterfaces for the operation of extended attributes and content of file. In the systemprototype, extended attributes can also ensure the integrity and scalability of file’ssemantic information, so that we can avoid loosing association in traditional file system.Also, by adding attached file to collect hot data, for target files, and designing the easilyacceptable view for users, we can improve the performance of retrieval.With some functional and performance tests, this prototype system proved that it hasimplemented the function of metadata extensions, file associations and file retrieval etc.Functional tests are used to check file-system’s normal function and operations forextended attributes. Performance tests are mainly used to check the amount of time used tobuilding index and retrieval. As the realization taking into account,comparing to normalfile system,user space file system framework is used in prototype system, which leads to almost20%performance reduction in the normal operations test. The results show that itis faster than third-party search tools in the document retrieval, and the performance isimproved a lot.The extra time overhead due to conversion of a documentis about10microseconds.