Implementation schemes for a crash recovery mechanism

Atomic actions represent a powerful tool for system structuring, controlling accesses to shared data and implementing backward error recovery techniques. In this paper properties of atomic actions are analyzed, specifically referring to crash recovery problems. In order to guarantee consistency of shared data in spite of crashes, programs implementing atomic actions must satisfy some constraints. Implementation schemes are developed in sequential and concurrent contexts. Different schemes are separately described, namely: single object and multiple objects actions, uniprocess and multiprocess actions. Practical problems that arise during the implementation of atomic actions are pointed out.