Nested atomic action and recoverability in distributed systems

Abstract

The nested atomic action is an appealing primitive to cope with the reliability issues peculiar to distributed systems, the uncertainty and the partial failure. This paper addresses the nested atomic action and the recoverability in distributed systems. This paper presents the system structuring scheme to support the nested atomic actions in distributed systems adopting the multilevel systems model. The scheme centers around the recovery data management scheme. Based upon the primitive regarding to the recovery data management, the necessary algorithms for the functions of the underlying interpreter are presented. The algorithms include those for establishment and discard of a recovery point, commit protocol, recovery, and remote call interface. The algorithms in coordination assure a computation executing over several nodes to be atomic in face of the failure. While the scheme seems abstract in a sense, the functionality of the design is shown through a typical example.