Home > Publications database > Strategien zur Kopplung und Datenreduktion bei der Online-Visualisierung von parallelen Simulationsrechnungen mit verteilter Datenhaltung |
Diploma Thesis | PreJuSER-26050 |
2002
Forschungszentrum Jülich GmbH Zentralbibliothek, Verlag
Jülich
Please use a persistent id in citations: http://hdl.handle.net/2128/2585
Report No.: Juel-4021
Abstract: Visualizing data just being calculated by a simulation program running an a remote computer is called online visualization. This method allows an immediate visualization and a direct control of simulation parameters (computational steering). Applying online visualization to large-scale applications an parallel supercomputers with distributed memory lead to two major problems: an one hand the huge amount of data which have to be transfered between simulation and visualization and an the other hand the distributed data management within the simulation program. The first Problem is tackled by a compression procedure an the basis of a wavelet transformation which is implemented in this thesis. This allows to decompose the data in different resolution steps and to transfer them progressively to the visualization. To overcome the second problem coupling strategies for the distributed data management are introduced, which take into account the parallel structure of the simulation program and ensure a reliable transfer of the decomposed data located an the single processors. The benchmarking of the applied compression techniques and coupling strategies with respect to the run time behavior and to the effects an a parallel simulation program was done by simplified models and validated by real time measurements. For that purpose a library $\textit{LVISIT}$ and a code generator $\textit{visitcg}$ were developed, which use the communication library $\textit{VISIT}$ and provide the coupling techniques introduced here together with a compression based an wavelet transformation.
The record appears in these collections: |