Evaluation of Software Dependability Based on Stability Test Data

This paper discusses a measurement-based approach to dependability evaluation of fault-tolerant, real-time software systems based on failure data collected from stability tests of an air traffic control system under development. Several dependability analysis techniques are illustrated with the data: parameter estimation, availability modeling of software from the task level, applications of the parameter estimation and model evaluation in assessing availability, identifying key problem areas, and predicting required test duration for achieving desired levels of availability, and quantification of relationships between software size, the number of faults, and failure rate for a software unit. Although most discussion is focused on a typical subsystem, Sector Suite, the discussed methodology is applicable to other subsystems and the system. The study demonstrates a promising approach to measuring and assessing software availability during the development phase, which has been increasingly demanded by the project management of developing large, critical systems.