Developing Inherently Resilient Software Against Soft-Errors Based on Algorithm Level Inherent Features

摘要

A potential peculiarity of software systems is that a large number of soft-errors are inherently derated (masked) at the software level. The rate of error-deration may depend on the type of algorithms and data structures used in the software. This paper investigates the effects of the underlying algorithms of programs on the rate of error-deration. Eight different benchmark programs were used in the study; each of them was implemented by four different algorithms, i.e. divide-and-conquer, dynamic, backtracking and branch-and-bound. About 10,000 errors were injected into each program in order to quantify and analyze the error-derating capabilities of different algorithm-designing-techniques. The results reveal that about 40.0% of errors in the dynamic algorithm are derated; this figure for backtracking, branch-and-bound and divide-and-conquer algorithms are 39.5%, 38.1% and 28.8%, respectively. These results can enable software designers and programmers to select the most efficient algorithms for developing inherently resilient programs. Furthermore, an analytical examination of the results using one-way ANOVA acknowledged the statistical significance of difference between the algorithm-designing-techniques in terms of resiliency at 95% level of confidence.