A Fault-list Generation Approach Based on Data Flow Analysis

Fault injection is an effective technique used to evaluate fault tolerance mechanisms.It accelerates the evaluation processing by injecting faults into the system.However,these faults are typically selected at random from the total fault space of the system.These selection has one significant limitation;that is,the no response problem,which leads inaccurate evaluation of the fault tolerance mechanisms.In this paper,we propose a fault list generation approach based on data flow analysis.Firstly,a Single Event Upset(SEU) fault model is proposed,which is composed of Fault-injection location,Fault-injection time,Fault type and Fault mask.Then Fault-injection location is calculated by live variables analysis,and Fault-injection time is calculated by reaching definitions analysis.Finally,this fault list generation approach is validated to be having an accelerated failure rate of 90% on the PowerPC8548 Processor and its Trace Simulator.