Software reliability enhancement against hardware transient errors using inherently reliable data structures

Abstract

Decreasing the scale of transistors and their voltages and exponential increase in the transistor counts have made the nowadays digital integrated circuits more susceptible to transient hardware errors (soft errors). One of the interesting features of software systems is that a considerable number of soft errors are inherently masked at software level. The likelihood of error masking (error deration) in the software may be influenced by the Algorithm, data structures and programming paradigms used in the software. One of the main research questions in this field of study is that how can software reliability be improved against soft errors without external redundancy and only by selecting appropriate software structures. This paper investigates the inherent effects of the underlying data structures on the rate of error deration and program reliability. To attain this goal, five different benchmark programs were implemented by four different data structures, i.e. Array, Binary-search Tree, One-way linked list and Two-way linked list; profiling experiments were performed on the benchmarks to identify those features of the data structures which affect the rate of error-deration. Then, in order to quantify and examine the inherent reliability of the data structures, about 5,600,000 faults were injected into the benchmark programs. The results show that about 53.95% of faults in the Array based programs are masked; this figure for Binary-search tree, One-way linked list and Two-way linked list are 40.16%, 44.02% and 42.73%, respectively. We found that Array and BST as two different data structures have the highest and lowest inherent reliability respectively. These findings enable the software developers to select the most reliable data structures for developing reliable programs without external redundancy.