Quantifying selective duplication in multithreaded parallel applications

Quantify the fault coverage and runtime costs of instruction-level and kernel-level selective duplication in multithreaded parallel applications.

Background

PaRID supports selective duplication at both the instruction level, by duplicating only vulnerable instructions, and the kernel level, by duplicating instructions only within vulnerable OpenMP kernels. The paper notes that, unlike in serial programs, evaluating the trade-off between fault coverage and runtime for these selective strategies in multithreaded parallel applications requires non-trivial analysis. The authors explicitly leave this quantification for future work.

References

Unlike serial programs, quantifying the fault coverage and runtime of selective duplication in multithreaded parallel applications requires non-trivial analysis. We consider this as our future work.

— Detecting Soft Errors in Parallel Software with LLM-tuned Instruction Duplication  (2609.19531 - Huang et al., 17 Sep 2026) in Section 5, subsection “Fault Coverage”