Comparing Static and Dynamic Weighted Software Coupling Metrics

TL;DR

This paper compares static and dynamic weighted coupling metrics in software systems, revealing surprising correlations and differences at class and package levels through four weeks of real-world data.

Contribution

It introduces the analysis of weighted dynamic coupling metrics and compares them to static metrics using empirical data from commercial software systems.

Findings

High correlation between static and dynamic weighted metrics.

Differences observed between class-level and package-level analyses.

Dynamic metrics provide additional insights into runtime behavior.

Abstract

Coupling metrics are an established way to measure software architecture quality with respect to modularity. Static coupling metrics are obtained from the source or compiled code of a program, while dynamic metrics use runtime data gathered e.g., by monitoring a system in production. We study \emph{weighted} dynamic coupling that takes into account how often a connection is executed during a system's run. We investigate the correlation between dynamic weighted metrics and their static counterparts. We use data collected from four different experiments, each monitoring production use of a commercial software system over a period of four weeks. We observe an unexpected level of correlation between the static and the weighted dynamic case as well as revealing differences between class- and package-level analyses.