Scheduling message processing for reducing rollback propagation
Yueqi Wang, W.K. Fuchs · 2003
The authors show that the probability of rollback propagation in a message-passing system can often be greatly reduced by reordering the processing of messages. Also, rollback propagation was measured for several parallel programs. A scheduling algorithm for message processing and its implementation for reducing rollback propagation are described. The algorithm incorporates a user-transparent prioritized scheme based on the run-time communication and checkpointing history. Communication trace-driven simulation for several parallel programs written in the Chare Kernel language demonstrated that the probability of rollback propagation can be reduced at the cost of slight additional performance degradation.>