Automatic recovery of a parallel stable file system

This article describes the automatic recovery algorithm and its implementation for the POOSS stable file system. The recoverable faults by our scheme include non-consecutive disk loss, single duplication loss, single duplication corruption, and duplication inconsistency. The recovery algorithm is implemented on the 100-node parallel machine (POOMA). One important feature of the recovery algorithm is that it is node-independent and can be executed simultaneously on different nodes, resulting in a highly parallel recovery procedure.