Methods and apparatus for process replication/recovery in a distributed system

摘要

A distributed computing system includes a number of computers, workstations or other computing machines interconnected by a network. A non-interactive process arriving in a host machine of the system is migrated for execution to at least two remote machines. For example, first and second executions of the process may be started on respective first and second remote machines. One of the first and second executions of the process is then used to provide an on-demand checkpoint for the other execution of the process in the event the other execution is terminated, such that an additional execution of the process can be started from the on-demand checkpoint. This on-demand checkpointing is augmented with periodic checkpointing performed on at least one of the multiple executions of the process. The period of the periodic checkpointing for a given execution of the process may be fixed without regard to the status of the on-demand checkpointing for that execution, or alternatively may be reset each time an on-demand checkpoint is taken for that execution.