ZURUECK HOCH VOR INHALT SUCHEN

» Back to overview
Proposing Institution

Institut für Informatik, Universität Köln
Project Manager

Charilaos Tzovas
Weyertal 121
50931 Köln
Abstract
Mesh partitioning is an indispensable tool for efficient parallel numerical simulations. Its goal is to achieve load balance while minimizing communication between the processes of the simulation. While established graph-based partitioning tools yield a high solution quality, their scalability is limited. Geometric approaches usually scale better, but their solution quality is comparably poor for non-trivial mesh topologies.In this project, we investigate a scalable version of k-means that is also adapted to yield balanced clusters. Balanced k-means constitutes the core of our new partitioning algorithm;it is complemented by space-filling curves for bootstrapping and graph-based local refinement for quality-oriented postprocessing.By combining geometric and graph-based methods, we address the aforementioned limitation of scalability and solution quality.

Impressum, Conny Wendler