Detecting distributed termination in the presence of node failure

Abstract

Distributed termination detection concerns detecting the termination of a distributed computation spread across a set of processing nodes. Many solutions have been proposed. However, most solutions have not addressed the problem of the failure of the processing node. In this paper, a termination detection scheme which tolerates node failure is proposed. It is both symmetric and distributed. The scheme can be applied to any kind of connection topology. It uses the FIFO property of a single transmission line to cope with system wide non-FIFO message transmission. Some of the information concerning termination detection is backed up for fault tolerant purposes. The “backing up” does not require the synchronisation of the processing nodes, and it does not interfere with the normal computation. Therefore, the overhead to achieve fault tolerance has been minimised. The scheme is derived using the invariant technique.