Characterization of the Impact of Soft Errors on Iterative Methods

Mutlu, Burcu O.; Kestor, Gokcen; Manzano, Joseph; Ünsal, Osman; Chatterjee, Samrat; Krishnamoorthy, Sriram

doi:10.1109/hipc.2018.00031

Cited by 5 publications

(7 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These ephemeral data structures are usually well-tuned for DRAM-based high-performance computing (HPC) systems [4,19,45,46]. Large-scale HPC systems are likely to be interrupted by failures because its components are not reliable [21,35,50]. Consequently, checkpointing is required for long-running HPC applications designed with the ephemeral data structures to provide failure recovery.…”

Section: Introductionmentioning

confidence: 99%

Compiler aided checkpointing using crash-consistent data structures in NVMM systems

Coy

Ren

et al. 2020

Proceedings of the 34th ACM International Conference on Supercomputing

View full text Add to dashboard Cite

Scientific applications use checkpointing for failure recovery. The existing checkpointing approaches were proposed for storing persistent states of applications as checkpoints in disk-based file systems via the block interface. As non-volatile main memory (NVMM) will be included in high-performance computing systems, storing the checkpoints in NVMM-based file systems can significantly waste the performance benefits of NVMM. This is because it underutilizes memory resources and it does not take advantage of the byte-addressability of NVMM. In this paper, we propose an NVMM-aware checkpointing approach, named NV-Checkpoint. It uses a compiler-aided technique to automatically generate multi-version data structures, which consist of both the persistent version of data stored in NVMM for failure recovery and the ephemeral version of data placed across DRAM and NVMM. Because of the byte-addressability of NVMM, any versions can be accessed via the memory interface. The multiple versions may share data that are not mutated during the program's execution to reduce data redundancy. NV-Checkpoint provides the same level of guarantee of failure recovery compared to the conventional checkpointing approaches proposed for file systems. Furthermore, its runtime system manages the layout of the data structures to reduce the number of writes to NVMM. It also manages the checkpointing frequency to reduce persistence overhead using machine learning models. Our experimental results with real-world scientific applications show that the performance of annotated programs with NV-Checkpoint using a hybrid of DRAM and NVMM matches the performance of best-effort handwritten versions. It achieves similar scalability as those with ephemeral data structures using only DRAM. It offers up to 121X speedup of execution time

show abstract

Section: Introductionmentioning

confidence: 99%

Compiler aided checkpointing using crash-consistent data structures in NVMM systems

Coy

Ren

et al. 2020

Proceedings of the 34th ACM International Conference on Supercomputing

View full text Add to dashboard Cite

show abstract

“…As discussed in the previous section, the data can be leveraged for further anaylsis and can facilitate future studies. Therefore, at the end of this study, we created a publicly available database called Iterative Method Injection Collection at https://github.com/pnnl/IMIC [58]. This database has all 1.75 million injection results with traces collected during this study.…”

Section: Imic Databasementioning

confidence: 99%

“…That is, for the n th iteration, we compare the injected vector value with the vectors at the {n-20 .. n+20} iterations. We compare the injected vector with the corresponding healthy vector's values over the course of iteration window, as any change introduced by the error can hinder the convergence, but also, by chance, it can help moving the execution to the right direction [58]. We calculated ℓ1 norms between each vector and the correct execution's corresponding vector range.…”

Section: • Iteration Percentage Calculated Asmentioning

confidence: 99%

“…Our error injection framework instruments the source code to simulate an error. The framework decides in where to inject the error based on the inputs provided by the user (similar to our previous works [46,58]). These inputs are: iteration number, statement number, vector name, position in the vector, list of bit positions to flip in the vector element.…”

Section: Error Injection Mechanismmentioning

confidence: 99%

“…This way, we make sure our injection won't be overwritten and can have a chance to affect the program state. Error behavior including such overwriting can be calculated mathematically from this set of experiments without additional experiments as shown in prior work [58]. We inject uniform random 1-bit, 2-bit, and 4-bit errors.…”

Section: Error Injection Mechanismmentioning

confidence: 99%

See 2 more Smart Citations

An extensive study on iterative solver resilience : characterization, detection and prediction

Mutlu

View full text Add to dashboard Cite

Soft errors caused by transient bit flips have the potential to significantly impactan applicalion's behavior. This has motivated the design of an array of techniques to detect, isolate, and correct soft errors using microarchitectural, architectural, compilationbased, or application-level techniques to minimize their impact on the executing application. The first step toward the design of good error detection/correction techniques involves an understanding of an application's vulnerability to soft errors. This work focuses on silent data e orruption's effects on iterative solvers and efforts to mitigate those effects. In this thesis, we first present the first comprehensive characterizalion of !he impact of soft errors on !he convergen ce characteris tics of six iterative methods using application-level fault injection. We analyze the impact of soft errors In terms of the type of error (single-vs multi-bit), the distribution and location of bits affected, the data structure and statement impacted, and varialion with time. We create a public access database with more than 1.5 million fault injection results. We then analyze the performance of soft error detection mechanisms and present the comparalive results. Molivated by our observations, we evaluate a machine-learning based detector that takes as features that are the runtime features observed by the individual detectors to arrive al their conclusions. Our evalualion demonstrates improved results over individual detectors. We then propase amachine learning based method to predict a program's error behavior to make fault injection studies more efficient. We demonstrate this method on asse ssing the performance of soft error detectors. We show that our method maintains 84% accuracy on average with up to 53% less cost. We also show, once a model is trained further fault injection tests would cost 10% of the expected full fault injection runs. “Soft errors” causados por cambios de estado transitorios en bits, tienen el potencial de impactar significativamente el comportamiento de una aplicación. Esto, ha motivado el diseño de una variedad de técnicas para detectar, aislar y corregir soft errors aplicadas a micro-arquitecturas, arquitecturas, tiempo de compilación y a nivel de aplicación para minimizar su impacto en la ejecución de una aplicación. El primer paso para diseñar una buna técnica de detección/corrección de errores, implica el conocimiento de las vulnerabilidades de la aplicación ante posibles soft errors. Este trabajo se centra en los efectos de la corrupción silenciosa de datos en soluciones iterativas, así como en los esfuerzos para mitigar esos efectos. En esta tesis, primeramente, presentamos la primera caracterización extensiva del impacto de soft errors sobre las características convergentes de seis métodos iterativos usando inyección de fallos a nivel de aplicación. Analizamos el impacto de los soft errors en términos del tipo de error (único vs múltiples-bits), de la distribución y posición de los bits afectados, las estructuras de datos, instrucciones afectadas y de las variaciones en el tiempo. Creamos una base de datos pública con más de 1.5 millones de resultados de inyección de fallos. Después, analizamos el desempeño de mecanismos de detección de soft errors actuales y presentamos los resultados de su comparación. Motivados por las observaciones de los resultados presentados, evaluamos un detector de soft errors basado en técnicas de machine learning que toma como entrada las características observadas en el tiempo de ejecución individual de los detectores anteriores al llegar a su conclusión. La evaluación de los resultados obtenidos muestra una mejora por sobre los detectores individualmente. Basados en estos resultados propusimos un método basado en machine learning para predecir el comportamiento de los errores en un programa con el fin de hacer el estudio de inyección de errores mas eficiente. Presentamos este método para evaluar el rendimiento de los detectores de soft errors. Demostramos que nuestro método mantiene una precisión del 84% en promedio con hasta un 53% de mejora en el tiempo de ejecución. También mostramos que una vez que un modelo ha sido entrenado, las pruebas de inyección de errores siguientes costarían 10% del tiempo esperado de ejecución.

show abstract