ABSTRACT
This paper shows the features and the stucture of the software quality measurement model developped by THOMSON-CSF/DSE, and arrangements made for its implementation and its exploitation.
The Electronic Systems Division (DSE) created within the THOMSON group in 1967 is responsible for the development of ground to air missile systems.
Because of the critical role played by on-board software with regard to the structure and functioning of such weapons systems, the Quality Assurance Department of DSE is heavily oriented towards achievement of quality objectives for embedded software. It has been working since 1988 to define software quality management procedures including the setting up of a programme aimed at measuring the quality of software produced within the Division.
The quality measurement model has been largely based upon Mc Call’s initial concepts, with the following important extensions:
– enhanced coverage of test and specification phases. Quality measurements must be made throughout the whole development process, from requirements specification through to qualification.
– coverage of production process quality in addition to product quality. This follows from the belief that the quality of any product depends both upon the intrinsic quality of the product, and the quality of the processes by which it has been produced.
– establishment of explicit relations between the factors, criteria and metrics and the phases of the development cycle during which measurements are taken.
– 163use of a model memory to take into account, during calculations for a specific phase, the results obtained during earlier development activities.
Values for each metric can, according to the specific nature, either be taken directly from a work product or else calculated from a weighted checklist.
This raw data is then processed in accordance with the following principles by which the objects of the model are quantified:
– thresholds are applied to measurements of terminal metrics,
– values of non-terminal objects are obtained by calculating a normalised heuristic,
– object values can either be interpreted directly from the numeric values, or by threshold assignment.
For effective analysis of the vast volumes of data handled by the model, and the efficient implementation of management actions responding to the results provided, the data must be synthesised for presentation to disparate parties.
Two classes of summary report were designed:
– Model Status Report, showing the quality levels of the factors and sub-factors of the model,
– Factor Status Report, illustrating the status through time.
In this and the above report, colour coding conventions are used to increase the readability of the reports by associating different colours to value intervals.
The quality measurement model described in this paper is currently undergoing pilot test evaluation within the Division.
We hope that this paper can contribute to the ongoing debate in the areas covered, and encourage the continued investigation into quantifiable software quality management.
