Not logged in.
Quick Search - Contribution
Contribution Details
| Type | Bachelor's Thesis |
| Scope | Discipline-based scholarship |
| Title | Orthogonal Recursive Bisection on the GPU for Accelerated Load Balancing in Large N-Body Simulations |
| Organization Unit | |
| Authors |
|
| Supervisors |
|
| Language |
|
| Institution | University of Zurich |
| Faculty | Faculty of Business, Economics and Informatics |
| Date | 2022 |
| Abstract Text | Large simulations with billions of particles are used to understand the universe. Commonly, the fast multipole method is employed to improve the runtime, where a space partitioning binary tree data structure is required. The structure is generated using the Orthogonal Recursive Bisection method and at the same time used as a load balancing strategy to leverage the hardware of super computers optimally. In this paper, a GPU accelerated version of ORB is proposed and implemented in the CUDA programming language. The implementation manages to maintain a consistent runtime during the entire execution of the ORB method, regardless of the increasing complexity as fragmentation grows with the tree depth and the number of leaf cells. Performance measurements showed a speedup by a factor of 5.4 over a fully parallelized and optimized CPU version. |
| PDF File |
Download
|
| Export |
BibTeX
|