Software method cuts error checking cost in parallel programs
Detecting Soft Errors in Parallel Software with LLM-tuned Instruction Duplication
Distributed, Parallel, and Cluster ComputingArtificial IntelligenceProgramming Languages
Summary
Soft errors can make computer programs give wrong answers, and checking for these errors slows programs down a lot. The authors created PaRID, a software tool that improves error checking in programs that run many tasks at once. It works by cleverly copying instructions and using large language models to tune how it protects the code. Their tests show PaRID greatly cuts the usual slowdown, making error detection faster while still finding all errors.
What this means in practice
- •For software engineers for high-performance computing: Improve reliability of multithreaded applications by reducing overhead of error detection during parallel execution.
- •For systems reliability teams: Enable faster, more efficient detection of soft errors in servers running large-scale parallel workloads to maintain system integrity.
Authors
Yafan Huang, Guanpeng Li
Abstract
We propose PaRID (PaRallel Instruction Duplication), a software-directed soft error detection framework that requires only compile-time effort for multithreading parallel programs. PaRID addresses two key challenges: supporting parallel programs with mixed serial and parallel regions and minimizing performance overhead without relying on costly dynamic profiling. It combines parallel-aware code transformation with LLM-tuned performance modeling, guided by eight generalizable findings from an offline characterization study, to enable fast soft error detection in parallel applications. Evaluation on NPB benchmarks shows that PaRID reduces protection overhead from 162.79% to 59.84% on average and achieves up to 5x speedup while maintaining full error detection effectiveness.