Exploring Traditional and Emerging Parallel Programming Models Using a Proxy Application

Abstract

Abstract-Parallel machines are becoming more complex with increasing core counts and more heterogeneous architectures. However, the commonly used parallel programming models, C/C++ with MPI and/or OpenMP, make it difficult to write source code that is easily tuned for many targets. Newer language approaches attempt to ease this burden by providing optimization features such as automatic load balancing, overlap of computation and communication, message-driven execution, and implicit data layout optimizations. In this paper, we compare several implementations of LULESH, a proxy application for shock hydrodynamics, to determine strengths and weaknesses of different programming models for parallel computation. We focus on four traditional (OpenMP, MPI, MPI+OpenMP, CUDA) and four emerging (Chapel, Charm++, Liszt, Loci) programming models. In evaluating these models, we focus on programmer productivity, performance and ease of applying optimizations.

The Mont-Blanc Prototype: An Alternative Approach for HPC Systems

“…Later, we compare Mont-Blanc against MareNostrum supercomputer on a subset of the applications to make a direct comparing between the two, in terms of performance and energy consumption. [14], [15] Electronic Structure BQCD [16] Quantum Chromodynamics MP2C Multi-Particle Collision Dynamics QuantumESPRESSO [17] Electronic Structure and Materials Modeling SMMP [18], [19], [20] Molecular Thermodynamics Alya [21], [22] Biomedical Mechanics COMD [23] Proxy for Molecular Dynamics LULESH [24], [25] Proxy for Hydrodynamics miniFE [26] Proxy for Finite Element Method…”

ACURDION: An adaptive clustering-based algorithm for tracing large-scale MPI applications

“…Lulesh is large enough to be more complex than traditional benchmarks, yet compact enough to support a large number of implementations [10]. We observe a slight diffusion in the communication pattern in Fig.…”

LULESH Programming Model and Performance Ports Overview

“…A brief classification of the less common models is presented, but details are not covered. For more information about most of these ports along with a look at the programmability, optimizability and performance portability of the LULESH implementations in these languages please see [10].…”

“…The same solution as used in the OpenMP codes are applied in CUDA. Some optimizations specific to the Fermi based GPU are contained in this version and the reader is referred to [10] for more details.…”

“…Hydrodynamics was chosen as one of the challenge problems because it consumes 27% of data center resources throughout DOD. Mini-apps like LULESH provide a smaller full featured program that allow for less time consuming exploration of various programming models and performance tuning techniques [10,9]. The successful lessons learned from the mini-app exploration can then be applied to a full application.…”