A Taxonomy to Enable Error Recovery and Correction in Software
Source: Northeastern University
Over the past several years, reliability research has largely used the following taxonomy of errors: Undetected Errors (sometimes referred to as Silent Data Corruption, or SDC), Detected Uncorrectable Errors (DUE), and Corrected Errors (CE). While this taxonomy is suitable to characterize hardware Error Detection And Correction (EDAC) techniques, it does not provide enough granularity to assess system-wide recovery techniques. Specifically, it does not provide the ability for architects to assess DUE severity, an ability which is crucial to evaluating many error tolerance schemes.