Towards a resilient shared memory concept for distributed persistent object systems

The abstraction of a centralized memory which has been successfully introduced in shared memory Multiprocessors has also been adopted for Distributed Shared Memory (DSM) in loosely coupled, network-based distributed systems as an adequate paradigm to handle shared state. In this paper a concept for a resilient distributed shared memory is elaborated which takes advantage of an advanced action model. The action model guarantees consistency of data in the presence of concurrency and failures by structuring the system activities and supporting damage assessment and recovery. It is argued that the concurrency control mechanism of actions can be exploited to guarantee consistency among multiple copies of an object. Additionally, an integrated mechanism is provided to increase availability in the action system by replication. The paper analyses the consequences of combining replication with the action system.