Demystifying the System Vulnerability Stack: Transient Fault Effects Across the Layers

Papadimitriou, George; Gizopoulos, Dimitris

doi:10.1109/isca52012.2021.00075

Cited by 42 publications

(11 citation statements)

References 67 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, evaluations on complete designs and large applications might involve unfeasible evaluation times. Other strategies, including software-based error propagation on GPUs, depend on the error models used, which might induce simplified evaluations and inaccurate analyses [28]. Thus, strategies to effectively characterize fault effects while providing an acceptable trade-off between performance and accuracy are still required.…”

Section: A Motivation and Related Workmentioning

confidence: 99%

“…Low-level hardware fault simulations (e.g., gate level) offer detailed insights but are time-consuming, while higher-level hardware or software simulations (e.g., microarchitectural or software) provide faster but less accurate results. Additionally, selecting appropriate fault models for simulation is critical to prevent inaccurate conclusions [42]. In contrast, physical fault injection using radiation exposes the system to a beam of ionizing particles, offering a means to estimate realistic error rates.…”

Section: A Motivation and Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Special Session: Reliability Assessment Recipes for DNN Accelerators

Ahmadilivani,

Bosio,

Deveautour

et al. 2024

2024 IEEE 42nd VLSI Test Symposium (VTS)

View full text Add to dashboard Cite

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

show abstract

Section: A Motivation and Related Workmentioning

confidence: 99%

Section: A Motivation and Related Workmentioning

confidence: 99%

Special Session: Reliability Assessment Recipes for DNN Accelerators

Ahmadilivani,

Bosio,

Deveautour

et al. 2024

2024 IEEE 42nd VLSI Test Symposium (VTS)

View full text Add to dashboard Cite

show abstract

“…Nevertheless, existing reliability optimization techniques for traditional edge computing can serve as a guide for improving the reliability in serverless edge computing scenarios. This is because both scenarios are susceptible to common failure types, including transient faults caused by electromagnetic interference and cosmic radiation [23] and bit errors during data communication [24]. In the literature, researchers have explored reliability-aware computation offloading and assignment [15], developed frameworks to minimize the block error probability during communication [16], devised strategies to tolerate transient faults and bit errors [17].…”

Section: Reliability Augmentationmentioning

confidence: 99%

REPFS: Reliability-Ensured Personalized Function Scheduling in Sustainable Serverless Edge Computing

Cao,

Weng

2024

IEEE Trans. Sustain. Comput.

View full text Add to dashboard Cite

In recent years, serverless edge computing has been widely employed in the deployments of Internet-of-things (IoT) applications. Despite considerable research efforts in this field, existing works fail to jointly consider essential factors such as energy, reliability, personalized user requirements, and stochastic application executions. This oversight results in an inefficient utilization of computation and communication resources within serverless edge computing networks, subsequently diminishing the profit of service providers and degrading the quality-ofexperience (QoE) of end users. In this paper, we explore the problem of reliability-ensured personalized function scheduling (REPFS) to jointly optimize the profit of service providers and the holistic QoE of end users in sustainable serverless edge computing. A personality-driven user QoE prediction method is first designed to accurately estimate the QoE of individual end users with differentiated personality types. Afterward, a deterministic function scheduling policy is developed on the problem-specific augmented non-dominated sorting genetic algorithm II (PSA-NSGA-II). Given the inherent uncertainty of application executions, a stochastic function scheduling strategy that can be easily parallelized for modern multicore scheduler platforms is also devised to accelerate solution generation for stochastic applications. Experimental results show that our deterministic function scheduling policy achieves 15% performance enhancement compared with representative multiobjective evolutionary algorithms B-NSGA-II, c-DPEA, DLS-MOEA, MOALO, MOEA/D, and MaOEA/IGD. Furthermore, our stochastic function scheduling strategy promotes the service profit by 78% and the holistic user QoE by 118% on average compared with the developed deterministic scheduling policy.

show abstract

“…Faults can be injected at different levels of abstraction, from low-level Register-Transfer Level (RTL) [32]- [34] to microarchitecture [35], [36] and software [19], [37]- [41]. Each level of abstraction provides the propagation probability of the injected fault to the output, measured in terms of Hardware, Architectural, Software, or Program Vulnerability Factor (HVF, AVF, SVF, PVF, respectively) [42], [43]. Fault injection assumes that the fault already occurred and provides the fault propagation through the system.…”

Section: Radiation Induced Errors In Dnnsmentioning

confidence: 99%

Characterizing a Neutron-Induced Fault Model for Deep Neural Networks

Santos

Kritikakou

Condia

et al. 2023

IEEE Trans. Nucl. Sci.

View full text Add to dashboard Cite

The reliability evaluation of Deep Neural Networks (DNNs) executed on Graphic Processing Units (GPUs) is a challenging problem since the hardware architecture is highly complex and the software frameworks are composed of many layers of abstraction. While software-level fault injection is a common and fast way to evaluate the reliability of complex applications, it may produce unrealistic results since it has limited access to the hardware resources and the adopted fault models may be too naive (i.e., single and double bit flip). Contrarily, physical fault injection with neutron beam provides realistic error rates but lacks fault propagation visibility. This paper proposes a characterization of the DNN fault model combining both neutron beam experiments and fault injection at software level. We exposed GPUs running General Matrix Multiplication (GEMM) and DNNs to beam neutrons to measure their error rate. On DNNs, we observe that the percentage of critical errors can be up to 61%, and show that ECC is ineffective in reducing critical errors. We then performed a complementary software-level fault injection, using fault models derived from RTL simulations. Our results show that by injecting complex fault models, the YOLOv3 misdetection rate is validated to be very close to the rate measured with beam experiments, which is 8.66× higher than the one measured with fault injection using only single-bit flips.

show abstract

Demystifying the System Vulnerability Stack: Transient Fault Effects Across the Layers

Cited by 42 publications

References 67 publications

Special Session: Reliability Assessment Recipes for DNN Accelerators

Special Session: Reliability Assessment Recipes for DNN Accelerators

REPFS: Reliability-Ensured Personalized Function Scheduling in Sustainable Serverless Edge Computing

Characterizing a Neutron-Induced Fault Model for Deep Neural Networks

Contact Info

Product

Resources

About