An Approach to Communication Reduced Parallel Multigrid

A straightforward parallelization of multigrid is known to be asymptotically optimal in the sense that all extra overhead for communication and synchronization is of the same or lower order than the arithmetic cost, when the problem size grows and the number of processors is kept fixed. Unfortunately, in many practical examples the amount of communication cannot be neglected so simply, and many algorithms have been suggested to alleviate this problem. We consider an approach to reduce communication based on ideas of Brandt and Diskin [2] and examine its possible gains for different parallel architectures. [2] Brandt, A., Diskin, B.: Multigrid Solvers on Decomposed Domains; Contemporary Mathematics, 157 (1994), 135 - 155

@article{id723,
  author = {Mohr, M. and R\"ude, U.},
  journal = {Zeitschrift f\"ur Angewandte Mathematik und Mechanik},
  language = {en},
  note = {Proceedings of the GAMM annual meeting 1998},
  number = {Suppl. 3},
  pages = {901{\textendash}902},
  title = {An Approach to Communication Reduced Parallel Multigrid},
  volume = {79},
  year = {1999},
}

%O Journal Article
%A Mohr, M.
%A Rüde, U.
%J Zeitschrift für Angewandte Mathematik und Mechanik
%G en
%O Proceedings of the GAMM annual meeting 1998
%N Suppl. 3
%P 901–902
%T An Approach to Communication Reduced Parallel Multigrid
%V 79
%D 1999