Gem5Panalyzer: A Light-weight tool for Early-stage Architectural Reliability Evaluation & Prediction
| Autor: | Xiaoxing Qiu, Hao Qiu, Semiu A. Olowogemo, Daniel B. Limbrick, Bor-Tyng Lin, William H. Robinson |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
010302 applied physics
020203 distributed computing Artificial neural network Computer science business.industry Deep learning 02 engineering and technology 01 natural sciences Extensibility Reliability engineering 0103 physical sciences 0202 electrical engineering electronic engineering information engineering qsort Artificial intelligence business |
| Zdroj: | MWSCAS |
| DOI: | 10.1109/mwscas48704.2020.9184536 |
| Popis: | Aggressive technology scaling boosts the embedded processor performance but deteriorates the system resilience to soft errors. To balance trade-offs between reliability and design overheads, we developed Gem5Panalyzer: a novel light-weight reliability evaluation tool built on top of the gem5 simulator. It computes vulnerability factors (VFs) of the programs running on ARMv7-like CPUs and constructs a dataset for machine learning studies. Gem5Panalyzer features satisfactory compatibility and extensibility. Another highlight of our tool is the integration of reliability and machine learning. We validated Gem5Panalyzer by evaluating the behavior of selected MiBench programs’ VFs. Additionally, a study using Neural Network and Gem5Panalyer to predict VFs shows that for the qsort application, the predictor performance has become saturated when the training set size reaches 43%. This finding attests the idea that predicting future VFs of large applications is possible based on a limited prior knowledge of vulnerability. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|