PRASE: An Approach for Program Reliability Analysis with Soft Errors

Abstract

Soft errors are emerging as a new challenge in computer applications. Current studies about soft errors mainly focus on the circuit and architecture level. Few works discuss the impact of soft errors on programs. This paper presents a novel approach named PRASE, which can analyze the reliability of a program with the effect of soft errors. Based on the simple probability theory and the corresponding assembly code of a program, we propose two models for analyzing the probabilities about error generation and error propagation. The analytical performance is increased significantly with the help of basic block analysis. The program¿s reliability is determined according to its actual execution paths. We propose a factor named PVF (program vulnerability factor), which represents the characteristic of program¿s vulnerability in the presence of soft errors. The experimental results show that the reliability of a program has a connection with its structure. Comparing with the traditional fault injection techniques, PRASE has the advantage of faster speed and lower price with more general results.