A Concept for Hybrid Fault Injection in Distributed Systems

Fault tolerance is an essential part in the design of distributed computer systems [2]. Thus, testing algorithms for such systems often involves fault injection (FI) experiments to evaluate if the specified fault hypothesis holds for the implementation, i.e., faults can be seen as part of the valid input data of a fault-tolerant system. Basically, the different methods of FI can be grouped into three main categories (see [5] for a comparison): Simulation-based FI [1], hardware FI [6], and software FI [4]. Hardware FI tends to be less intrusive where timing is concerned than software FI and thus preferable. It is primarily used on the chip-level, however. Software FI, on the other hand, while allowing fault injection on the algorithmic level, considerably affects the timing behavior of the system and thus is not suitable for typical networked embedded systems with real-time constraints. In this paper, we present a concept of a FI framework for distributed systems1that is based on a hybrid hardwaresoftware based method of fault injection. Additional to the hybrid approach, we aim for a global fault model, where the conditions for the injection of a specific fault may depend on the internal states of all nodes of the tested distributed system, similar to the system proposed with the Loki [3] architecture. The improvement with regard to the Loki environment will be the implementation of a hardware-assisted FI mechanism that improves the timing properties of the FI system.