File Schemes for Fault Tolerant Support in Heterogeneous Distributed Storage Systems

摘要

One important piece of system software for clusters is the parallel file system. Two main problems can be found in current parallel file systems and parallel I/O libraries for clusters. One is that they do not use standard servers, thus it is very difficult to use these systems in heterogeneous environments. Other is that with multiple servers running long time is more probably a fault that blocks clients. There are not a clear solution that integrate, inside the parallel file system itself, the fault tolerance support needed in a flexible way. We propose the use of existing network file servers at the server end of a parallel file system to offer a parallel file system for heterogeneous clusters. By integrating redundant techniques into the parallel file system, is possible to define a fault tolerance support from partition level to file level. This paper describes the design of the fault tolerance support for the next version of the Expand Parallel File System. and a preliminar evaluation that has been made in Linux clusters of the first prototype.