A Systematic Approach to Automatically Generate Multiple Semantically Equivalent Program Versions

Abstract

Classic methods to overcome software faults include design diversity that involves creating multiple versions of an application. However, design diverse techniques typically require a staggering investment of time and manpower. There is also no guarantee that the multiple versions are correct or equivalent. This paper presents a novel approach that addresses the above problems, by automatically producing multiple, semantically equivalent copies for a given array/loop-based application. The copies, when used within the framework of common design diverse techniques, provide a high degree of software fault tolerance at practically no additional cost. In this paper, we also apply our automated version generation approach to detect the occurrence of soft errors during the execution of an application.

This work is supported in part by NSF grants # 0720645 , # 0702519 and support from the Gigascale Systems Research Focus Center, one of the five research centers funded under SRCs Focus Center Research Program. The authors would like to thank the anonymous reviewers for their helpful remarks. The authors would like to thank Seung Woo Son and Shiva Prasad Kasiviswanathan for their suggestions. Finally, the authors would like to thank, our shepherd, Dr. Erhard Plödereder who helped finalize the paper.