Estimation of software release times based on a queueing model for software testing

Abstract

The development o f large-scale software systems is a complex and expensive process. The key to successful software projects is planning and estimation o f time frames for each phase o f the lifecycle. Testing plays a very important role in software lifecycle and constitutes almost 40% o f the total development costs. In the literature, several software reliability models have been proposed to assess the reliability o f software from either specified parameters or from software-error generated data. These models assume the impractical notion o f perfect debugging, and also assume that no new faults are introduced in the event o f imperfect debugging. Also, these models do not take into account the characteristics o f the software and the process o f its development. In this thesis, we propose a queueing model based on re-entrant lines to depict the process o f software modules undergoing testing/debugging, verification and validation, and quality assurance tests before being accepted for use. In the proposed model for software testing, servers denote the test teams and parts denote the software modules to be tested. The model takes into account the structure o f the software, the individual modules being distinguished by their criticality in the mission, the complexity -of design and implementation, and their usage in operational field from profiles. Reentrancy o f modules is attributed to the notion o f imperfect debugging and that new faults can be introduced in the event o f imperfect debugging. Using the re-entrant line model for software testing, bounds on test times are obtained by considering the state transitions for a general class o f modules and solving a linear programming model. Scheduling o f software modules for test at each process step yields the constraints for the linear program. From the bounds on test times, bounds on product release times are obtained. We resort to simulation modeling to obtain the mean test times for modules. Different sets o f simulation parameters are used based on the structure o f the software, with the individual modules categorised based on their complexity, criticality and usage. The performance bounds obtained through the linear programs formulated and the mean performance values obtained using detailed simulation are useful in the planning o f test activities and estimating software release times in software development efforts for large projects.