Concepts for efficient multigrid implementation on SUPRENUM-like architectures (Q804270)

The implementation of multigrid methods on parallel computers with distributed memory is discussed. To minimize the exchange of certain intermediate results between the parallel processes various possibilities are offered. The main idea is the use of an optimized multicolour relaxation scheme combined with an adapted agglomeration technique. Performance estimations for the SUPRENUM supercomputer are presented for the example of the Poisson equation to be solved in non-orthogonal boundary-fitted coordinates using a 9-point difference operator with variable coefficients.