SoRel is a tool for Software Reliability analysis and prediction. It is composed of two parts allowing reliability growth tests and application of reliability growth models. It allows two kinds of failure data processing (inter-failure data and number of failures per unit of time, i.e. failure intensity data). The main measures that can be evaluated are: the mean time to next failure (or MTTF), the intensity function, the cumulative number of failures and the residual failure rate of the software.
Four reliability growth tests are available: the arithmetical mean (for inter-failure times data), the Laplace test , the Kendall test and the Spearmann test (for both inter-failure and failure intensity data). These tests allow periods of time during which reliability is increasing or decreasing to be identified in order to apply reliability growth models on data exhibiting trends in accordance with their assumptions (which generally leads to better predictions).
Four reliability growth models are implemented allowing different kinds of behavior to be modeled:
- a decreasing failure rate tending to zero when time tends to infinity: the exponential model (Goel-Okumoto) which is an NHPP and the Littlewood-Verrall model which is a failure rate model,
- a decreasing failure rate tending to a non zero value: the hyperexponential model (Kanoun-Laprie) which is an NHPP model
- an increasing failure rate followed by a decreasing failure rate: S-Shaped model (Yamada et al.).
Model validation criteria are the Kolmogorov-Smirnov distance, the prequential likelihood and the relative residue when the input data consists of inter-failure times and the residue (mean error) when failure intensity data is used. A model can be analysed according to its retrodictive capability (its ability to reproduce the observed behavior of the software) and predictive capability (its ability to predict future behavior of the software, from the observed failure data).
The results are available into two forms:
- immediately on the screen (numerical results and curves),
- in the form of files which can be read and used by other Macintosh applications (Excel, Word…), numerical results and curves can thus be used directly for publications and reports.
The program is modular and new reliability growth tests and models can easily be added. It is written in Pascal and has the advantages of multiple window management since it runs in a Macintosh environment.The human/machine interface has been denoted special attention.
For documentation look at:
K. Kanoun, M. Kaaniche, J.C. Laprie, "Qualitative and Quantitative Reliability Assessment", IEEE Software, Vol. 14 (2). pp. 77-87, March-April 1997.
For more information please contact: Karama Kanoun (kanoun@laas.fr)