The Software Implemented Fault Injection Study Based On Single Event Effects

With the development of spaceflight technology, the reliability of onboard systems becomes more and more important. The fault tolerance technology is an important approach to improve the reliability of onboard computer systems, sequencely, its correctness and efficiency need to be ensured by verification. Comparing with the other two methods of verification of fault tolerance including the modeling and simulation testing technology and the field data analysis, the fault injection technology is more popular and more effective because of its advantages in the aspect of convenience, faster speed and lower price with more general results about the application reliability. This dissertation makes a systematic study on the two key problems of the fault injection technology, which are the fault propagation process and the fault model. Fault model is the most important part of the fault injection experiment.This paper focuses on the fault propagation process in program on the level of assembly languages. We analyze the Single Event Fault's influence on computer systems and propose a probabilistic model of error generation. Then we discuss the generation method of fault flow diagram of basic block, execution path and program in detail. Based on data flow graph, the effect of fault in memory cell is quantitatively described by the utilization of the definition of Risk Probablity. The method of selecting the risk path and critical data by Risk Probablity is presented as well, which provide foudantion for the generation of fault model.Fault model is an important factor which influences the experiment efficency and the accuracy of experiment results. In this dissertation, two key problems related to fault model are deeply studied, including the development of Single Event Fault Pattern and the design of fault list. We propose the Single Event Fault Pattern which consists of fault injection location, fault injection time, fault type and fault mask, establishing the fault space of fault injection experiment. Based on risk path and critical data, we present a method of equivalent set partition to generate fault list. Our fault list generation method solves the problems of fault explosion and low pertinence of the fault list generated by random sampling and stratified sampling. Besides, this method is suitable for software implemented fault injection technology with a high system failure rate since the fault injection location and the fault injection time are especially considered.Finally, the design and implementation of SimFi fault injector is introduced, which is a software implemented fault injection tool based on SimpleScalar.The simulation in our case study based on SimFi demonstrates the effectiveness of our fault list generation method.