Software Program-Level Reliability Modeling and Estimation

Abstract

In order to estimate the reliability at the software program- level while accounting for the knowledge from the underlying hardware layers, this chapter presents different reliability estimation models that are developed at different levels of granularity, i.e., instruction and function/task level. Since each system layer may employ distinct reliability optimization techniques that can operate at either the instruction or function/task level, it is important to devise reliability models for the appropriate granularity adapted to these optimization techniques. For example, a metric at an instruction granularity will be useful for enabling reliability optimization during compilation. However, at the system software layer the notion of function/task is more appropriate. A key challenge to develop efficient software program-level reliability models is to identify important hardware- and software-level parameters that affect the reliability of a software program executing on unreliable hardware. For this, the analysis of Chap. 3 is important to be considered, i.e., the knowledge of critical and noncritical instructions, spatial and temporal vulnerability, and error masking can be leveraged to develop accurate software program-level reliability models. These models are then used to analyze the reliability properties of different applications at the instruction and function granularity.