Cross-layer system reliability assessment framework for hardware faults

Abstract

System reliability estimation during early design phases facilitates informed decisions for the integration of effective protection mechanisms against different classes of hardware faults. When not all system abstraction layers (technology, circuit, microarchitecture, software) are factored in such an estimation model, the delivered reliability reports must be excessively pessimistic and thus lead to unacceptably expensive, over-designed systems. We propose a scalable, cross-layer methodology and supporting suite of tools for accurate but fast estimations of computing systems reliability. The backbone of the methodology is a component-based Bayesian model, which effectively calculates system reliability based on the masking probabilities of individual hardware and software components considering their complex interactions. Our detailed experimental evaluation for different technologies, microarchitectures, and benchmarks demonstrates that the proposed model delivers very accurate reliability estimations (FIT rates) compared to statistically significant but slow fault injection campaigns at the microarchitecture level.Peer ReviewedPostprint (author's final draft

SyRA: Early System Reliability Analysis for Cross-Layer Soft Errors Resilience in Memory Arrays of Microprocessor Systems

“…A first attempt of using a Bayesian cross-layer reliability model to estimate the system level failure in time was presented by the authors of this paper in [53]. Despite providing preliminary good results, the model proposed in the paper is simplified especially when it comes to the way the different layers are interfaced.…”

“…This allows us to speed up the analysis and therefore to cope with the complexity of the simulation of the full software stack. SyRA extends a preliminary attempt to use Bayesian networks for crosslayer reliability analysis [53]. Compared to [53], SyRA proposes a revised model that changes the way technological parameters are modeled in order to support the estimation of different reliability metrics including: Architecture Vulnerability Factor (AVF), Failures In Time (FIT) rate, and Executions Per Failure (EPF).…”

“…SyRA extends a preliminary attempt to use Bayesian networks for crosslayer reliability analysis [53]. Compared to [53], SyRA proposes a revised model that changes the way technological parameters are modeled in order to support the estimation of different reliability metrics including: Architecture Vulnerability Factor (AVF), Failures In Time (FIT) rate, and Executions Per Failure (EPF). In particular, the last metric enables the designer to tradeoff reliability and performance in a single measure providing a valuable tool to optimize a computing system.…”

“…In particular, the last metric enables the designer to tradeoff reliability and performance in a single measure providing a valuable tool to optimize a computing system. Moreover, SyRA optimizes the model of the interface between components at different layers in order to cope with the complexity of the analysis in case of very complex applications, thus overcoming the scalability issues of [53] when considering large realistic software applications.…”

Kernel and layer vulnerability factor to evaluate object detection reliability in GPUs

“…However, their overhead may be unacceptable for a real-time safety-critical system. Previous works have been focusing on selective hardening for central processing units (CPUs), seeking the best point on performance and reliability trade-off with interesting and promising results [23][24][25]. In this study, we propose an efficient software hardening strategy for HOG and YOLO.…”