FORTRESS: A System to Support Fail-Aware Real-Time Applications
Christof W. Fetzer, Flaviu Cristian · 1997
Fortress is a support system for designing and implementing fault-tolerant distributed real-time systems that use commercial of the shelf (COTS) components. The main problem we address in Fortress is that services cannot always provide their standard properties due the possibility of missed deadlines, dropped messages and process crashes. Fortress allows clients to detect when a service cannot provide its standard semantics anymore due to unmasked failures. A service is fail-aware if it maintains an indicator that allows its clients to determine if the service provides it standard semantics or some predefined exception semantics. Fortress provides fail-aware clock synchronization, membership and atomic broadcast services. Indicators allow a fail-safe application to switch the system to a safe state in case not all failures can be masked.