Method and apparatus for checkpointing in computer system

摘要

A computer system monitors inter-process communications and performs a synchronous (global) checkpointing for processes that belong to a checkpoint group. The system also performs a local checkpointing at respectively arbitrary times within each process. When a fault occurs, a validity of each checkpoint is examined in accordance with monitoring results of inter-process communications related to the fault process. If the most recent checkpoint of the fault process is valid, only the fault process is rolled back. If it is invalid, all processes of the checkpoint group are rolled back to the global checkpoint, or each of the processes are rolled back to each optimum (valid) checkpoint.