ECE PhD student, Guanpeng Li, and his advisor, Dr. Karthik Pattabiraman, received the Best Paper Award Runner Up at the 48th IEEE/IFIP International Conference on Dependable Systems and Networks (DSN), 2018. The conference took place from June 25-28 in Luxembourg City.
Their paper is titled Modeling Soft-Error Propagation in Programs, and was one of three papers to be nominated for the award out of nearly 300 submissions. The DSN conference is considered to be the most prestigious venue in the field of dependable computing.
The paper proposes a new technique, Trident, to estimate the error resilience of a program through static and dynamic analysis methods. Traditionally, error resilience was estimated through the use of fault injection experiments, which are very time-consuming. Consequently, researchers have proposed alternate techniques to estimate the error resilience of a program without performing fault injections. However, those techniques suffer from significant inaccuracies. The paper from the ECE team showed that it is possible to achieve high accuracy without the use of fault injections, thereby significantly accelerating the resilience estimation process. Says Dr. Pattabiraman about the paper, “Trident represents a fundamental advance in the way dependable systems are designed. In the past, practitioners were loathe to integrate resilience techniques into the software development process as such techniques took too long to run. In contrast, Trident is almost two orders of magnitude faster than traditional techniques, and can hence be integrated into the software development process easily.”
The work was done in collaboration with researchers from Nvidia Research in the USA, and was funded in part through a Strategic Grant from the NSERC. The ECE team hopes to integrate Trident into a production compiler in the near future. The Trident code is freely available here.