next up previous
Next: Checkpointing and Process Migration. Up: Implementing the BSP Model Previous: The Bayanihan Master-Worker Based

Adaptive Parallelism and Fault-Tolerance

  Since no communications occur while these work packets are running, we can rerun work packets or run several copies of the same work object at a time, resolving any redundancies safely at the end of the superstep. This makes it possible to implement adaptive parallelism and fault-tolerance mechanisms.


Luis Sarmenta