Distributed File System

With the development of the society, data in all kind of fields increases at an amazing speed,especially the coming of the Internet Age. Meanwhile, huge amount of information needs to bestored in the digital way. As a result, how to store and manage the digitized information isbecoming the focus concerned. In the 70's and 80's, these data were stored in the expensivemiddle or mini computers. System administrators must periodically back up the data, which isexpensive and easy to cause mal-manipulation. Recently, however, people have turned to studythe distributed system along with the prompt development of the PC hardware as well as network.In the meantime, Open-Source Operating System (OSOS) has been widely used. Consequently,many distributed systems based on OSOS have emerged to store and manage data. These systemsare generally composed of high-performance PC interconnected through the high-speed networkequipments, which could provide huge concurrent data stream, high price/performance as well ashigh availability.Distributed and Parallel File System (DPFS) discussed in this thesis is the core of thedistribute system. It can manage the data among the servers in the system dynamically to obtainthe optimal performance for the system. Developed based on Linux, DPFS is specifically for thedistributed and parallel server system differing from the common client/server model.The thesis firstly describes the development of the data storage and the trend of new demand.Then it introduces effects that the DPFS exerts on the data storage, and factors that should beconsidered while designing DPFS, goals that the design of DPFS should be reached to meet theperformance requirements for distributed server system. After that, the thesis presents logicstructure of DPFS , duplicate table model as well as the consistency of duplicate table. It alsoanalyzes the effect that DPFS exerts on the reliability of the system and presents the results incomparison with EXT3 while reading and writing. At last, it makes expectation for the futureresearch for the distributed and parallel file system .The thesis emphasizes on the design of duplicate table and its implementation algorithmsincluding the physical and logical structure , the cache's mechanism and the interface which itprovides . On the basis of duplicate manage , It puts forward an dynamical duplicate managementmodel, such as the strategy of creating and deleting duplicate. Then it shows the consistency ofduplicate which involves the consistency of duplicate table and the consistency of master and itsduplicate as well as an application that periodically checks the duplicate in order to keep theconsistency of the master and its duplicates.