Programming guidelines for improving software resiliency against soft-errors without performance overhead

摘要

A large percentage of soft-errors which alters a computer system are inherently derated (masked) due to the structural features of software. The higher usage of error-derating structures in a computer program leads to the higher resiliency. Computer programming style, as a principle phase of software development, has inherent and hidden effects on the program resiliency. Hence, improving the resiliency of a computer system without using external redundancy and hardware modification and only by making a simple modification in the program source-code and structure is one of the main issues in this field of study. This paper investigates the inherent side effects of a subset of programming structures and styles on the rate of error-deration and consequently program resiliency; To attain this goal, four different benchmark programs were implemented by different programming structures. Profiling experiments were performed to identify the inherent effects of the programming structures on the rate of program vulnerability against soft-errors and the error-deration; then, in order to examine the effects of the programming structures on the program resiliency, about 10,080,000 faults were injected into the benchmark programs. The results reveal that using x++ operator (rather than x = x + 1 and x+ = 1), do-while structure (rather than for structure) and inline functions (rather than normal and recursive functions) improves the program resiliency about 8%. This finding can enable software programmers to develop highly resilient programs without using external redundancy.