Investigating the Impact of High-Level Software Design on Low-Level Hardware Fault Resilience

Silent Data Corruptions (SDCs) have become an insurmountable issue that threatens the system reliability. General strategies for protecting programs from SDCs, such as dual modular redundancy, incur intolerable overheads. Another strategy is Algorithm-Based Fault Tolerance which is highly bounded to the specific algorithm and hard to generalize. In this study, we find different implementations of the same algorithm may lead to very different SDC probabilities. We conduct a characterization study to quantify the differences and investigate the root causes. The insights we derive could help and guide the developers in software engineering domain to design programs that is naturally resilient.