Fault Modeling of Extreme Scale Applications Using Machine Learning

Abstract

Faults are commonplace in large scale systems. These systems experience a variety of faults such as transient, permanent and intermittent. Multi-bit faults are typically not corrected by the hardware resulting in an error. This paper attempts to answer an important question: Given a multi-bit fault in main memory, will it result in an application errorand hence a recovery algorithm should be invoked-or can it be safely ignored? We propose an application fault modeling methodology to answer this question. Given a fault signature (a set of attributes comprising of system and application state), we use machine learning to create a model which predicts whether a multibit permanent/transient main memory fault will likely result in error. We present the design elements such as the fault injection methodology for covering important data structures, the application and system attributes which should be used for learning the model, the supervised learning algorithms (and potentially ensembles), and important metrics. We use three applications-NWChem, LULESH and SVM-as examples for demonstrating the effectiveness of the proposed fault modeling methodology.

Tuning EASY-Backfilling Queues

“…Some works use collaborative filtering to colocate tasks in clouds by estimating application interference [30]. Others are closer to the application level and use binary classification to distinguish benign memory faults from application errors in order to execute recovery algorithms (see [31] for instance).…”

Empirical Mathematical Model of Microprocessor Sensitivity and Early Prediction to Proton and Neutron Radiation-Induced Soft Errors

“…ML-based methods have been used in several ways. Vishnu et al [18] proposed a combination of eight ML algorithms, to assess the impact of multi-bit memory errors on HPC applications and to compare their predictions with the fault-injection results. In [19], the authors used linearregression techniques to backtrack the propagation of soft errors through processes dependent on many-core processor systems.…”

Using Machine Learning Techniques to Evaluate Multicore Soft Error Reliability

“…While this approach focus on the reduction the of total time to complete the fault injection campaigns, our approach aims at correlating large subsets of application profiles and architecture characteristics with fault injection results in order to pinpoint the most relevant parameters/traces on the target system. Vishnu et al [12] evaluate the impact of multi-bit memory errors, both permanent and transient, on HPC applications. This work considers eight different ML algorithms (e.g., support vector machines, k-Nearest neighbors, three distinct decision trees), comparing their predictions (i.e., the error probability) with the ground-truth (i.e., fault injection results).…”