Multiple criteria:

1. When you go from one machine running the code to more than one 2. Any system that may experience failures and detection of such failures and recovery is desirable 3. Most distributed systems due to the failure scenarios inherent in such systems.