The long-term goal of this project was to quantify visual computing systems, i.e., to assess, model, and ultimatelypredict important characteristics that have a substantial impact on user experience. We set out to use these models for algorithmic optimizations and cost savings, e.g., to maintain interactive frame rates during visual exploration of large data sets. While the focus during the first funding period was on runtime performance, this was extended to other performance metrics including throughput, perceived image quality, and energy consumption in the second period. In particular, the trade-offs between different metrics were a key point of investigation. Another major goal was comparing different visual computing methods in this regard. We identified two key visual computing scenarios that should be used as applications scenarios due to their current and projected relevance in the future: scientific visualization in high-performance environments and virtual and augmented reality.
How can we extend performance models from the computer architecture to deal with heterogeneous visual computing architectures and interactive applications?
How can we find the adequate level of abstraction and the relevant parameters for making quantitative performance predictions?
Can our framework support an application to adapt to variable loads and conditions in real time?
Can we extend the model to the adaptive algorithms and the perceptual metrics investigated in other research projects?
How can our models be used to give guarantees for minimal frame rates or maximal interaction latencies for specific data sets?
How can uncertainty be dealt with in terms of both the measurements and the predicted outcome?
Fig. 2: Experiment comparing inline, in transit and our hybrid configuration using a Cloverleaf3D simulation and Cinema database generation.
FOR SCIENTISTS
Projects
People
Publications
Graduate School
Equal Opportunity
FOR PUPILS
PRESS AND MEDIA