Aggregation AMG for Distributed Systems Suffering from Large Message Numbers

In iterative solution procedures for problems in science and engineering, AMG methods with inexpensive computation of the hierarchy of coarse-grid operators are a good choice to solve systems of linear equations where accurate solutions of these systems are not needed. In this contribution we demonstrate that the parallel performance of tins kind of algorithm is significantly improved if they are applied in combination with the Smoothed Aggregation approach, since this reduces the number of communication events. The resulting hybrid algorithms are particularly beneficial on systems where the number of messages limits the performance.