Experiments in indirect fault injection with open source and industrial software

Sammanfattning: Software faultinjection is a technique in which faults are injected into aprogram and the response of the program is observed. Faultinjection can be used to measure the robustness of the program aswell as to find faults in the program, and indirectly contributesto increased robustness. The idea behind software fault injectionis that the better the system handles the faults, the more robustthe system is. There are different ways of injecting faults, forexample, by changing a variable value to a random value or bychanging the source code to mimic programmer errors. The thesispresents an overview of fault injection in hardware and software.The thesis deals with a special case of fault injection, i.e.,indirect fault injection. This means that the faults are injectedinto one module and the response is observed in another module thatcommunicates with the first one. The thesis presents twoexperiments designed to measure the effect of the fault model usedwhen faults are injected using the indirect fault injection method.The first experiment is conducted on open source software. Theresult from the experiment was not entirely conclusive, but thereare indications that the fault model does matter, but this needs tobe further examined. Therefore, a second experiment is designed andpresented. The second experiment is conducted on larger, industrialsoftware. The goals of both experiments are to find out whether ornot the results of fault injection are affected by how the injectedfaults are generated. The second experiment shows the feasibilityof using fault injection in industrial strength software. Thethesis concludes with the proposal for a PhD thesis on a suite ofdifferent experiments.