A System Model for Distributed Real-time Systems

This report presents a system model for a class of distributed real-time systems. The goal is to assist the design of fault tolerance protocols such as membership agreement. The system model contains a description of the node, network, processing in respective sub-models. It also contains a failure model that describes the failures that can plausibly occur in the system. These failures can then be addressed by the fault tolerance protocol. The report contributes with a taxonomy by which failures can be described. The resulting failure model is affected by the model of the system and by the model of how processing is done in the system. The class of system is assumed to be strongly partitioned which provides a high degree of error containment for real-time processes executed in the same node and also for processes executed on different nodes. The smallest unit of failure is therefore the process. The system model uses a broadcast communication network similar to Flexray, i.e. it supports both time-triggered and event-triggered communication. The class of system is chosen based on the requirements of safety-critical applications such as x-by-wire. The processing model for the system is presented in which operation is divided into sequentially executed primitive operations, called CDR-operations (Compute and Distribute Result operations). A CDR-operation involves a producer process which computes a result and distributes the result via broadcast communication to consumer processes. Failures of CDR-operations are characterised by four aspects: type, symmetry, detectability and persistence; depending on which system component that is faulty. We compare our definitions of failure types with communication errors according to IEC 61784-3