NVBitFI

About NVBitFI

This tutorial targets to provide a consise instruction on using NVBitFI to perform fault injection experiments. NVBitFI is a fault injectior based on NVBit framework, which is a NVIDIA binary instrumentation tool. Why should we use NVBitFI? There are three reasons:

• No Need to Modify Code: Compared with LLFI-GPU, NVBitFI injects faults into SASS code level (i.e. NVIDIA GPU assembly code), so it does not require program source code and does not affect any instruction scheduling or register allocation.

• Very Very Fast: Compared with simulation-based fault injectors (e.g. GPGPU-SODA and GPGPU-SIM), NVBitFI is very fast (at GPU speed, billions of insts per sec) so it is not constrained to small kernels.

• GPU Arch-Agnostic: In addition, NVBitFI works for all recent NVIDIA GPU architectures including Kepler, Maxwell, Pascal, Volta, and Ampere.

Detailed descriptions see NVBitFI paper [1].

NVBitFI Installation

Fault Parameter that NVBitFI Supports

To Use NVBitFI

Contributing

This document is written by Yafan Huang. Besides, this document has not been iterated yet, so some descriptions may be a bit confused. If you have any questions after reading this document, please feel free to email him at yafan-huang@uiowa.edu.

References