Fault tolerant PVFS2 based on data replication

Aggregating the capacity and bandwidth of the commodity disks in the nodes of a cluster provides cost effective and high performance storage systems. Nevertheless, this strategy could be a feasible approach only if the mean time to failure of disks and nodes is faced. The number of failures increases with the nodes and it is especially important in parallel file systems, like PVFS, because having a file striped over server disks increases the probability of failures. This work proposes a strategy to include data replication in the second version of PVFS in order to provide fault tolerance. We also analyze the performance of the implementation of this approach.