JIT-Compilation for Interactive Scientific Visualization
Fraunhofer IGD, TU Darmstadt, Fraunhoferstr. 5, Germany 64283, Darmstadt, Hessen
24th WSCG Conference on Computer Graphics, Visualization and Computer Vision (WSCG), 2016
@article{mueller2016jit,
title={JIT-Compilation for Interactive Scientific Visualization},
author={Mueller-Roemer, JS and Mueller, Johannes and Altenhofen, C},
year={2016}
}
Due to the proliferation of mobile devices and cloud computing, remote simulation and visualization have become increasingly important. In order to reduce bandwidth and (de)serialization costs, and to improve mobile battery life, we examine the performance and bandwidth benefits of using an optimizing query compiler for remote postprocessing of interactive and in-situ simulations. We conduct a detailed analysis of streaming performance for interactive simulations. By evaluating pre-compiled expressions and only sending one calculated field instead of the raw simulation results, we reduce the amount of data transmitted over the network by up to 2/3 for our test cases. A CPU and a GPU version of the query compiler are implemented and evaluated. The latter is used to additionally reduce PCIe bus bandwidth costs and provides an improvement of over 70% relative to the CPU implementation when using a GPU-based simulation back-end.
May 7, 2016 by hgpu