Design And Implementation Of Network Coding Based Distributed File System

As electronic and information technology develops, new information is produced and processed in an unprecedented speed, along with that the network propagate information even faster. This shapes an urgent storage demand for huge amount of information. However local information storage imposes several limitations on functions and performance, such as low reliability, poor mobility and inconvenience for file sharing. So people begin to move their focus to new solution like network storage. Distributed file system is a basic ingredient of network storage, and its research has been emphasized by research community and industry.Distributed file systems should usually provide data storage with high reliability and good scalability, which is achieved by specific design of data placement methods in the systems' storage level. Some existing mechanisms include replication and erasure coding, each of which has its special objective and limitation. Research in this thesis is by using network coding to divide file into segments first and chunks further and encode them to be stored in the distributed environment. And network coding solutions to design and implementation are also shown in this thesis. The distributed file system in the thesis provides a uniform name space and conforms to location independence and transparency. It also supports basic file and directory operations and keeps the internal implementation transparent to users. The system adopts a centralized architecture, which is comprised by an index server, storage nodes and clients. The index server maintains load balance of the whole system through network coding mechanisms and repairs failed node when it is found. When a new node joins in, the index server will issue a re-balance process to provide the system with good scalability.This thesis firstly summarizes current research status of distributed file system and its system functions and classification. After an introduction to commonly used data placement and redundancy strategy, the network coding solution to data placement is shown. Later design and implementation details and solutions to load balance and node recovery are discussed. And the experiment design and performance evaluation results are given lastly.