Logo image
Model-Led Optimisation of a Geometric Multigrid Application
Conference proceeding

Model-Led Optimisation of a Geometric Multigrid Application

Richard A. Bunt, Simon J. Pennycook, Stephen A. Jarvis, B. Leigh Lapworth and Y. K. Ho
2013 IEEE 10th International Conference on High Performance Computing and Communications & 2013 IEEE International Conference on Embedded and Ubiquitous Computing, pp.742-753
11/2013

Abstract

Analytical models Computational modeling fluid dynamics Hardware high performance computing Libraries modelling Optimization performance analysis Program processors scientific computing
This paper details the construction of an analytical performance model of HYDRA, a production nonlinear multigrid solver used by Rolls-Royce for computational fluid dynamics simulations. The model captures both the computational behaviour of HYDRA's key subroutines and the behaviour of its proprietary communication library, OPlus, with an absolute error consistently under 16% on up to 384 cores of an Intel X5650-based commodity cluster. We demonstrate how a performance model can be used to highlight performance bottlenecks and unexpected communication behaviours, thereby guiding code optimisation efforts. Informed by model predictions, we implement an optimisation in OPlus that decreases the communication and synchronisation time by up to 3.01 times and consequently improves total application performance by 1.41 times.

Metrics

1 Record Views

Details

Logo image

Usage Policy