Exploratory Performance Analysis and Tuning of Parallel Interactive Volume Visualization on Large Displays

Abstract

We present an exploratory approach to performance analysis and tuning of interactive parallel volume visualization for large displays. While traditional approaches target non-interactive applications and focus on separate specialized views for post-mortem performance analysis, we show metrics from the GPU and volume ray casting together with the volume visualization and allow users to interact with both of them simultaneously. With this, users can explore the data set together with the corresponding metrics to investigate both the visual and the performance impact of different parameter settings jointly, like camera position, sampling density, or acceleration technique. In particular, this supports parameter tuning by providing the user not only with timings and quality measures, but also internal metrics from the GPU and the ray caster that help to understand the connection between parameter settings and their induced outcome. We demonstrate the usage and utility of our approach for performance analysis and tuning at the example of distributed volume ray casting for a high-resolution powerwall with the goal to achieve interactive frame rates with the best possible image quality.