For architectures with a single
multi-point communication
link (a bus)
K failures to tolerate => each operation is scheduled
K+1 times:
- The replica with the shortest completion time is the
main replica
It sends
its results to each replica of each successor
operation
(except when the result is known
locally)
- The failure of the main replica is detected by all the remaining
replicas thanks to the broadcast on the bus
The
remaining replica with the shortest completion time is chosen as the
new main replica
