A Language-Driven Tool for Fault Injection in Distributed Systems
GRID '05 Proceedings of the 6th IEEE/ACM International Workshop on Grid Computing
| Hi-index | 0.00 |
The availability of inexpensive powerful microprocessors leads to increasing deployment of those electronic devices in ever new application areas. Currently, the automotive industry considers the replacement of mechanical or hydraulic implementations of safety-critical automotive systems (e.g., braking, steering) by electronic counterparts (so-called "by-wire systems") for safety, comfort, and cost reasons.In order to remain operational in the presence of faults, these kinds of systems are built as fault-tolerant distributed real-time systems consisting of interconnected control units. To assure the correct operation of the fault tolerance mechanisms, software implemented fault injection provides low cost and easy to control techniques to test the system under faulty conditions.In this paper we propose a distributed software implemented fault injection framework based on the mobile agent approach. Software agents are designed to utilize the real-time system's global time and messages to trigger the fault injection experiments. We introduce a lightweight agent implementation language to model the fault injection and the concerned system resources, agent migration and logging of the fault injection experiments.We argue the feasibility of our approach by applying it to a "drive-by-wire" application. Here, incorporating the fault injection software in a mobile agent provides a clear separation of the fault injection software from the application software. Thus, the fault injection code is easily exchangeable -- providing means to inject different faults -- and can be removed smoothly from the production system after the test run is completed.