Abstract: Most safety-critical real-time applications integrate mechanisms for fault-tolerance. These mechanisms are often based on application-specific solutions (e.g. solutions depending on the semantics of used data), or depend on the run-time support functionalities (e.g. scheduling algorithm, network protocol) or depend on the use of specialized hardware (e.g. stable memory, replicated bus). In order to relieve application designers from the implementation of fault-tolerance mechanisms, to ease application evolution, and to reduce development costs, we describe in this paper a flexible environment for the design and execution of distributed safety-critical hard real-time applications, i.e. applications with strict timeliness requirements. Our approach is based on the definition of a generic task model, which is used to build both application tasks and tasks of the underlying run-time support. This model allows to achieve automatic off-line replication of application tasks to tolerate faults and detect errors. Replication relies on the definition of a set of transformation schemes applied on tasks' structure (active, semi-active, passive and temporal replication schemes). A minimal and predictable execution mechanism completes our approach. Its internal structure simplifies the integration of the run-time support costs into the application feasibility test, used to check respect of application strict timeliness requirements.

Keywords: real-time, safety-critical, automatic replication, predictable execution, Hades

report (gziped postscript, 118K).