3-D visualization of dynamic runtime structures

Abstract

Continued development and maintenance of software requires understanding its design and behavior. Software at runtime creates a complex network of call--callee relationships that are hard to determine but that developers need to understand to optimize software performance. Existing tools typically focus on static aspects (e.g., Structure101 or SonarQube), or they are difficult to use and require high expertise (e.g., software profiling tools). Unfortunately, these dependencies are hard to derive from static code analysis: For one, static analysis will reveal potential call--callee relationships not actual ones. Second, they are often difficult to detect, since information systems today increasingly use abstraction patterns and code injection, which obscures runtime behavior. In this paper, we present our efforts towards accessible and informative means of visualizing software runtime processes. We designed a novel visualization approach that utilizes a hierarchical and interactive 3-D city layout based on force-directed graphs to display the runtime structure of an application. This promises to reduce the time and effort invested in debugging programming errors or in finding bottlenecks of software performance. Our approach extends the city metaphor for translating programmatic relationships into accessible 3D visualizations. With the identified goals and constraints in mind, we designed a novel visual debugging system, which maps programming code structures to 3D city layouts based on force-directed graphs. Exploration of the animated visualization allows the user to investigate not only the static relationships of large software projects but also its dynamic runtime behavior. We conducted a formative evaluation of the approach with a preliminary version of a prototype. In a series of six interviews with experts in software development and dynamic analysis, we were able to confirm that the approach is useful and supports identifying bottlenecks. The interviews raised and prioritized potential future improvements, several of which we implemented into the final version of our prototype.