Efficient implementation of fault-tolerant data structures in embedded control software.

This paper presents a methodology and small software library which is intended to reduce the impact of transient data errors that may affect the software executing on commercial-of-the-shelf (COTS) embedded processors. The methodology involves duplication of data in disparate areas of memory (referred to as “mirror arrays”), and the complexity of the processing required to manage these areas is hidden by means of a library exporting new basic data types. Results are reported from three case studies in which the library was employed (a matrix multiplication program, a list-sorting program and a real-time control application): the findings obtained suggest that the methodology is highly effective in the presence of memory errors, the code changes required in order to use the library are very limited, and the impact on code readability is minimal.