Algebraic multigrid on a dragonfly network: First experiences on a cray XC30

Abstract

The Cray XC30 represents the first appearance of the dragonfly interconnect topology in a product from a major HPC vendor. The question of how well applications perform on such a machine naturally arises. We consider the performance of an algebraic multigrid solver on an XC30 and develop a performance model for its solve cycle.We use this model to both analyze its performance and guide data redistribution at runtime aimed at improving it by trading messages for increased computation. The performance modeling results demonstrate the ability of the dragonfly interconnect to avoid network contention, but speedups when using the redistribution scheme were enough to raise questions about the ability of the dragonfly topology to handle very communication-intensive applications.