College of Information Science & Technology University of Nebraska-Omaha, U.S.

---

Abstract—The purpose of this study is to compare the fitting (goodness-of-fit) and prediction capabilities of three reliability models using the failure data of five popular open source software (OSS) products. The failure data are modeled by Weibull and two other Non Homogenous Poisson Process (NHPP) models (Yamada S-Shaped and Schneidewind). The OSS products considered are Eclipse, Apache HTTP Server 2, Firefox, MPlayer OS X, and ClamWin Free Antivirus. Weibull is chosen due to its popularity in lifetime and its flexibility in modeling various distributions. On the other hand, among many software reliability models, the NHPP models are prevalent. The goodness-of-fit is based on the entire failure data collected. Prediction is accomplished by estimating the models parameters based on partial failure history and then applying the estimates to the entire time span for which failure data is collected. The outcomes show that a reliability model that fits the failure data well may not necessarily be a decent forecaster of future failure patterns.

Index Terms—Software reliability growth model, Nonhomogeneous poisson process (NHPP), Open source software (OSS), Prequential likelihood ratio (PLR), Weibull distribution.

[PDF]

Cite: Cobra Rahmani, Azad Azadmanesh and Lotfollah Najjar, "A Comparative Analysis of Open Source Software Reliability," Journal of Software vol. 5, no. 12, pp. 1384-1394, 2010.