Fitting of Software Error and Reliability Models to Field Failure Data

Abstract

The quantitative prediction and measurement of software reliability is of vital importance in the development of high quality cost effective software. Many software reliability models have been postulated in the literature [11], however few have been applied to field data. A model based upon the assumption that the failure rate of the software is proportional to the number of residual software errors leads to a constant failure rate and an exponential reliability function, [1]. The model contains two constants: the proportionality constant K and the initial (total) number of errors ET.The constants K and ET can be estimated during early design by comparison of the present project with historical data. During the integration test phase, a more accurate determination of the model parameters can be obtained by using simulator test data as if it were operational failure data. The simulator data is collected at two different points in the integration test phase and the two parameters can be determined from moment estimator formulas [9].The more powerful maximum likelihood method can also be employed to obtain point and interval estimates [3]. It is also possible to use least squares methods to obtain parameter estimates which is the simplest method and provides insight into the analysis of the data [12].This paper utilizes a set of development and file data taken on 16 different software developments [10] as a vehicle to study the ease of calculation and the correspondence of the three methods of parameter estimation. The sensitivity of the reliability predictions to parameter changes are studied and compared with field results. Additional theoretical studies plus comparisons between estimates and field measurements are needed to determine the optimum method of parameter estimation.The results show that if data is carefully collected, software reliability models are practical and yield useful results. These can serve as one measure to help in choosing among competitive designs and as a gauge of when to terminate the integration test phase.