Predictive analysis of a hydrodynamics application on large-scale CMP clusters

Abstract

We present the development of a predictive performance model for the high-performance computing code Hydra, a hydrodynamics benchmark developed and maintained by the United Kingdom Atomic Weapons Establishment (AWE). The developed model elucidates the parallel computation of Hydra, with which it is possible to predict its run-time and scaling performance on varying large-scale chip multiprocessor (CMP) clusters. A key feature of the model is its granularity; with the model we are able to separate the contributing costs, including computation, point-topoint communications, collectives, message buffering and message synchronisation. The predictions are validated on two contrasting large-scale HPC systems, an AMD Opteron/ InfiniBand cluster and an IBM BlueGene/P, both of which are located at the Lawrence Livermore National Laboratory (LLNL) in the US. We validate the model on up to 2,048 cores, where it achieves a > 85% accuracy in weak-scaling studies. We also demonstrate use of the model in exposing the increasing costs of collectives for this application, and also the influence of node density on network accesses, therefore highlighting the impact of machine choice when running this hydrodynamics application at scale.

Performance Modelling of Magnetohydrodynamics Codes

“…The authors show that a model can be built that is able to accurately predict the run-time of an application on a variety of architectures, whilst taking a minimal set of input parameters. This approach has successfully been used by others, including Davis et al [14] and Sundaram-Stukel et al [15], and is the basis of the approach taken in this paper.…”

Relative Performance Projection on Arm Architectures

“…The choices and approximations made to obtain these metrics are different in each model and result in differences in precision and speed. Some analytical approaches choose to reduce the problem by being application-dependent [4]. However, our model can explore different applications as we want to characterize diverse behavior in our applicative workload.…”

Optimisation of Patch Distribution Strategies for AMR Applications

“…Using system measurements provides an effective way to ensure work is evenly balanced [12], and the optimisations presented in Section 7 make use of a set of simple measures of system performance to develop an efficient patch distribution scheme. Our work differs from previous research by building on work in performance modelling [13,14], using measured parameters to estimate the optimal distribution of patches.…”