An efficient vulnerability-driven method for hardening a program against soft-error using genetic algorithm

Soft-errors are one of the major causes of software failures. Restricted reliability-improvement and undesirable performance-overhead are the main shortcomings of the state-of-the-art software-based methods to detect and recover soft-errors in a program. One of the main questions in this area of study is that which sections of the program, as the vulnerable sections, need to be duplicated against soft-errors? We propose a software-based method to tolerate soft-errors, as selective-replication, which precisely identifies and hardens the most vulnerable blocks of a program. Using the genetic algorithm (GA), the proposed method takes the dynamic behavior of the programs into consideration to identify the most vulnerable blocks. The results of fault-injection experiments show that, with about 30% duplication and about 24% performance-overhead, the proposed method leads to 82% error-detection coverage. Furthermore, the proposed method can be used to improve the efficiency of the statistical fault injection (SFI) methods which are used to evaluate the error coverage of a technique or reliability of a program; the injection space in a program is reduced about 30% by avoiding the fault injection in the derating-blocks which were identified by the proposed method.