Visualization of large three-dimensional data sets on consumer hardware requires pre-processing into hierarchical volume image data to reduce required data transfers onto GPUs, making the rendering process more efficient. This processing is typically an I/O-bound problem and can thus benefit from high bandwidth storage found in modern High Performance Computing (HPC) systems.

In this Bachelor thesis, a serial application for generating hierarchical volume image data will be extended to exploit multiple levels of parallelism on HPC hardware. In order to use the aggregated I/O bandwidth of multiple nodes, multi-process parallelism needs to be added to the application, e.g., using MPI. Additionally, thread-based parallelism will be required to better exploit the resources of a single compute node.

Envisioned Tasks

1. Investigation of suitable parallelisation strategies
2. Implementation of the proposed parallelisation strategy
3. Validation and analysis of the proposed solution
   1. Speedup and parallel efficiency
   2. Efficiency of the use of the I/O subsystem
4. Documentation of implementation and results in written form

For this work, basic knowledge of parallel programming in C/C++ will be required.

The language can be either German or English.

Contact

Jan Frenzel

jan.frenzel@tu-dresden.de