DECAF-FSEFI: A Fine-grained, Accountable, Flexible, and Efficient Soft Error Fault Injection Framework for Profiling Application Vulnerability

Resilient computation has been an emerging topic in the field of high-performance computing (HPC) for several years. In particular, studies show that tolerating faults on leadershipclass supercomputers (such as exascale supercomputers) is expected to be one of the main challenges. In this paper, we utilize dynamic binary instrumentation and virtual machine based fault injection to emulate soft errors and study the soft errors impact on the behavior of applications. We propose DECAF-FSEFI, a fine-grained, accountable, flexible, and efficient soft error fault injection framework built on top of QEMU. DECAF-FSEFI provides just-in-time fault injection, fault propagation trace, and flexible fault injection interfaces. In the case study, we demonstrate the usage of DECAF-FSEFI on fault injection experiments. While armed with so many features, the experiments illustrate that DECAF-FSEFI only introduces 2.48x performance overhead in the worst case and almost 0 overhead in the best case. Keywords—soft error; MPI; fault injection; resilience; vulnerability; High Performance Computing.