Statistics – Computation
Scientific paper
May 2004
adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2004agusmsm54a..04t&link_type=abstract
American Geophysical Union, Spring Meeting 2004, abstract #SM54A-04
Statistics
Computation
2753 Numerical Modeling
Scientific paper
Field line and stream line tracing is required in various space physics applications, such as the coupling of the global magnetosphere and inner magnetosphere models, the coupling of the solar energetic particle and heliosphere models, or the modeling of comets, where the multispecies chemical equations are solved along stream lines of a steady state solution obtained with single fluid MHD model. Tracing a vector field is an inherently serial process, which is difficult to parallelize. This is especially true when the data corresponding to the vector field is distributed over a large number of processors. We designed algorithms for the various applications, which scale well to a large number of processors. In the first algorithm the computational domain is divided into blocks. Each block is on a single processor. The algorithm folows the vector field inside the blocks, and calculates a mapping of the block surfaces. The blocks communicate the values at the coinciding surfaces, and the results are interpolated. Finally all block surfaces are defined and values inside the blocks are obtained. In the second algorithm all processors start integrating along the vector field inside the accessible volume. When the field line leaves the local subdomain, the position and other information is stored in a buffer. Periodically the processors exchange the buffers, and continue integration of the field lines until they reach a boundary. At that point the results are sent back to the originating processor. Efficiency is achieved by a careful phasing of computation and communication. In the third algorithm the results of a steady state simulation are stored on a hard drive. The vector field is contained in blocks. All processors read in all the grid and vector field data and the stream lines are integrated in parallel. If a stream line enters a block, which has already been integrated, the results can be interpolated. By a clever ordering of the blocks the execution speed can be increased dramatically. The communication occurs via file in an asynchronous manner. Disk access efficiency is improved by caching the data in memory.
de Zeeuw D.
Monostori G.
Tóth Géza
No associations
LandOfFree
Parallel field line and stream line tracing algorithms for space physics applications does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with Parallel field line and stream line tracing algorithms for space physics applications, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Parallel field line and stream line tracing algorithms for space physics applications will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-1179593