Checkpointing is defined as saving the current state of a run (parameter settings, contents of grid variables, and other relevant information) to a file. At a later time, this run can then be restarted from that state by recovering all the data from the checkpoint file.
Cactus checkpointing and recovery methods are provided by thorns. In general, these thorns decide how often to generate a checkpoint. They also register their recovery routines with the flesh; these recovery routines may then be called during initialisation of a subsequent run to perform the recovery of the state of the run. Such a recovery is requested by setting a parameter in the parameter file.
See Chapter C2.8 for details of how to create your own checkpointing and recovery methods.