Real-time software metrics

Abstract

This article describes the software metrics analysis of 10 releases of a real-time telephone switching system developed by a German telecommunications firm. A metrics program that computes the standard software complexity metrics plus information flow metrics was developed. The releases of the real-time software satisfy published laws of software evolution, e.g., continuing change, increasing entropy, and total change are not uniform over the changed modules. The data also support Harrison and Cook's program maintenance decision model. The data suggest use of the standard deviation of the changes in Halstead's V between releases as a threshold for their model. Factor analysis grouped the 18 complexity metrics into four dimensions of complexity: two from standard software complexity metrics (size and control flow) and two from information flow (global and resource variables referenced and global and resource variables changed). The information flow metrics characterize the complexity of the real-time parts of the software better than the standard software complexity metrics, e.g., Halstead's software science, lines of code, or McCabe's cyclomatic complexity.