2021/10/18 by Philipp Samfaß, Samfass, Philipp, Tobias Weinzierl +5
Computer Science · Engineering · #FOS: Computer and information sciences #Formal Methods in Verification #Hardware Architecture (cs.AR) #Mathematical Software (cs.MS) #Radiation Effects in Electronics #Software Engineering (cs.SE) #Software Reliability and Analysis Research
paper · pdf · doi:10.48550/arxiv.2110.15804
openalex publication_date 2021/10/18 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Resilient algorithms in high-performance computing are subject to rigorous\nnon-functional constraints. Resiliency must not increase the runtime, memory\nfootprint or I/O demands too significantly. We propose a task-based soft error\ndetection scheme that relies on error criteria per task outcome. They formalise\nhow ``dubious'' an outcome is, i.e. how likely it contains an error. Our whole\nsimulation is replicated once, forming two teams of MPI ranks that share their\ntask results. Thus, ideally each team handles only around half of the workload.\nIf a task yields large error criteria values, i.e.~is dubious, we compute the\ntask redundantly and compare the outcomes. Whenever they disagree, the task\nresult with a lower error likeliness is accepted. We obtain a self-healing,\nresilient algorithm which can compensate silent floating-point errors without a\nsignificant performance, I/O or memory footprint penalty. Case studies however\nsuggest that a careful, domain-specific tailoring of the error criteria remains\nessential.\n