Bibliographic Metadata

Title
Towards RBF Interpolation on Heterogeneous HPC Systems
AuthorHaase, Gundolf ; Dirk, Martin ; Günter, Offner
Published in
Large Scale Scientific Computing (LSSC'15) / Lirkov, I.; Margenov, S.; Wasniewski, J., Heidelberg, 2015, page 182-190
Published2015
LanguageEnglish
SeriesLNCS ; 9374
Document typeArticle in a collected edition
Keywords (EN)radial basis functions / FGP / shared memory / two-level
ISBN978-3-319-26520-9
URNurn:nbn:at:at-ubg:3-4160 Persistent Identifier (URN)
Restriction-Information
 The work is publicly available
Files
Towards RBF Interpolation on Heterogeneous HPC Systems [0.29 mb]
Links
Reference
Classification
Abstract (English)

We present a general approach for the parallelization of the interpolation with radial basis functions (RBF) on distributed memory systems, which might use various shared memory hardware as accelerator for the local subtasks involved. The calculation of an interpolant in general requires a global dense system to be solved. Iterative methods need appropriate preconditioning to achieve reasonable iteration counts. For the shared memory approach we use a special Krylov subspace method, namely the FGP algorithm. Addressing the distributed task we start with a simple block-Jacobi iteration with each block solved in parallel. Adding a coarse representation leads to a two-level block-Jacobi iteration with much better iteration counts and a wider applicability.

Notice
Stats
The PDF-Document has been downloaded 34 times.