Evaluating and Enhancing Robustness of Deep Recommendation Systems Against Hardware Errors

TL;DR

This paper systematically studies the robustness of deep recommendation systems against hardware errors, introduces an error injection framework, and evaluates mitigation techniques like activation clipping to improve system resilience.

Contribution

It is the first to analyze DRS robustness against hardware errors, develop Terrorch for error injection, and evaluate mitigation strategies such as activation clipping.

Findings

Activation clipping recovers up to 30% of degraded AUC-ROC scores.

DRS robustness varies with model parameters and input characteristics.

Error mitigation methods can significantly improve system resilience.

Abstract

Deep recommendation systems (DRS) heavily depend on specialized HPC hardware and accelerators to optimize energy, efficiency, and recommendation quality. Despite the growing number of hardware errors observed in large-scale fleet systems where DRS are deployed, the robustness of DRS has been largely overlooked. This paper presents the first systematic study of DRS robustness against hardware errors. We develop Terrorch, a user-friendly, efficient and flexible error injection framework on top of the widely-used PyTorch. We evaluate a wide range of models and datasets and observe that the DRS robustness against hardware errors is influenced by various factors from model parameters to input characteristics. We also explore 3 error mitigation methods including algorithm based fault tolerance (ABFT), activation clipping and selective bit protection (SBP). We find that applying activation clipping can recover up to 30% of the degraded AUC-ROC score, making it a promising mitigation method.