A mechanism of building a reliable parallel file system with unreliable nodes

This paper discusses the main problems when building a reliable parallel file system with unreliable nodes. And then, a new model with high-availability is proposed to solve these problems. In this model, two effective ways are used to implement this mechanism. First, a new node status transfer model is proposed to trace nodes' status and keep their data list consistently. Second, a new data protection and checksumming model is designed to protect the stored data. Finally, the tests results and analysis are presented in the aspects of function, performance and scalability