Fault tolerance lessons applied to parallel computing
Johannes Maier · 2002
The author describes an approach to fault-tolerant parallel computing which is based on experiences with the most successful fault-tolerant software-the transaction processing systems. The algorithms presented have less routine overhead and faster recovery than most preceding approaches. In the Pact (parallel actions) parallel programming environment, fault tolerance is provided that is fully user transparent in an easy-to-use parallel programming model with a simple synchronization model on virtual shared data.>