ZOFI: Zero-Overhead Fault Injection Tool for Fast Transient Fault Coverage Analysis (1906.09390v4)

Abstract: The experimental evaluation of fault-tolerance studies relies on tools that inject errors while programs are running, and then monitor the execution and the output for faulty execution. In particular, the established methodology in software-based transient-fault reliability studies, involves running each workload hundreds or thousands of times, injecting a random bit-flip in the process. The majority of such studies rely on custom-built fault-injection tools that are based on either a modified processor simulator, or a code instrumentation framework. Such tools are non-trivial to develop, and are usually orders of magnitude slower than native execution. In this paper we present ZOFI, a novel timing-based fault-injection tool that is aimed at being used in fault-coverage studies for transient faults. ZOFI is a zero-overhead tool, meaning that the analyzed workload runs at native speed. This is orders-of-magnitude faster compared to common approaches that are designed around simulators or code instrumentation tools. ZOFI is free software and is available at https://github.com/vporpo/zofi.