Dynamic reconfiguration of CSP programs for fault tolerance

In a distributed computation being performed by a network of communicating processes, failure of a process due to the failure of its host node can cause the entire computation to be aborted. The author proposes a scheme to make a distributed program resilient to the failure of one of its constituent processes. The distributed computation is completed despite the failure of a process. The scheme is for CSP programs and allows nondeterminism within a process. In CSP, the process name is used in input/output commands. Since synchronous communication is used, if a process specified in the input/output command of a process P does not execute a matching output/input command, P might get blocked. In the proposed scheme, if a process fails, another process starts executing on a backup node from the last checkpoint (CP) of the failed process. Programmed exception handling is used to ensure proper recovery and fault tolerance.<>