An MPI-CUDA implementation of an improved Roe method for two-layer shallow water systems

Asunción, Marc de la; Mantas, José M.; Castro, Manuel J.; Fernández-Nieto, Enrique D.

doi:10.1016/j.jpdc.2011.07.012

Cited by 22 publications

(5 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The main data structures are kept on the GPU, but there are no details as to how this was engineered. Similar results for unstructured meshes are later seen in [26].…”

Section: Related Worksupporting

confidence: 82%

Acceleration of a Python-Based Tsunami Modelling Application via CUDA and OpenHMPP

Weng

Strazdins

2014

2014 IEEE International Parallel &Amp; Distributed Processing Symposium Workshops

View full text Add to dashboard Cite

Modern graphics processing units (GPUs) have become powerful and cost-effective computing platforms. Parallel programming standards (e.g. CUDA) and directive-based programming standards (like OpenHMPP and OpenACC) are available to harness this tremendous computing power to tackle large-scale modelling and simulation in scientific areas. ANUGA is a tsunami modelling application which is based on unstructured triangular meshes and implemented in Python/C. This paper explores issues in porting and optimizing a Python/C-based unstructured mesh application to GPUs. Two paradigms are compared: CUDA via the PyCUDA API, involving writing GPU kernels, and OpenHMPP, involving adding directives to C code. In either case, the 'naive' approach of transferring unstructured mesh data to the GPU for each kernel resulted in an actual slowdown over single core performance on a CPU. Profiling results confirmed that this is due to data transfer times of the device to/from the host, even though all individual kernels achieved a good speedup. This necessitated an advanced approach, where all key data structures are mirrored on the host and the device.For both paradigms, this in turn involved converting all code updating these data structures to CUDA (or directiveaugmented C, in the case of OpenHMPP). Furthermore, in the case of CUDA, the porting can no longer be done incrementally: all changes must be made in a single step. For debugging, this makes identifying which kernel(s) that have introduced bugs very difficult. To alleviate this, we adopted the relative debugging technique to the host-device context. Here, when in debugging mode, the mirrored data structures are updated upon each step on both the host (using the original serial code) and the device, with any discrepancy being immediately detected. We present a generic Python-based implementation of this technique.With this approach, the CUDA version achieved 28× speedup, and the OpenHMPP achieved 16×. The main optimization of unstructured mesh rearrangement to achieve coalesced memory access patterns contributed to 10% of the former. In terms of productivity, however, OpenHMPP achieved significantly better speedup per hour of programming effort.

show abstract

“…The main data structures are kept on the GPU, but there are no details as to how this was engineered. Similar results for unstructured meshes are later seen in [26].…”

Section: Related Worksupporting

confidence: 82%

Acceleration of a Python-Based Tsunami Modelling Application via CUDA and OpenHMPP

Weng

Strazdins

2014

2014 IEEE International Parallel &Amp; Distributed Processing Symposium Workshops

View full text Add to dashboard Cite

show abstract

“…While a similar (simple) strategy has proven to be effective in other parallelizations and applications (e.g. [19], [20], [3]), the speedups here achieved were not quite positive, probably due to the excessive use of computationally inactive threads and overuse of global memory. At the contrary, the following approach has given more positive results and can be considered as a starting point for more sophisticated applications.…”

Section: A Whole Space Strategymentioning

confidence: 79%

“…[1], [2], [3]). In fact, the fast development of parallel tools has allowed the use of numerical simulations based on High Performance Computing (HPC) techniques for solving complex equation systems which govern the dynamics of real complex phenomena, in which scientists can efficiently study the modelling of, for instance, a lava flow [4], fire spreading [5] or pyroclastic flows [6].…”

Section: Introductionmentioning

confidence: 99%

CUDA Dynamic Active Thread List Strategy to Accelerate Debris Flow Simulations

Filippone

Spataro

D'Ambrosio

et al. 2015

2015 23rd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing

View full text Add to dashboard Cite

Cellular Automata represent a formal frame for dynamical systems which evolve on the base of local interactions. We here present first results of the CUDA parallelization of the SCIDDICA S3-hex Complex Cellular Automata model for simulating debris flows. In particular, a first strategy for the parallelization of the model is based on a straightforward one thread -one cell approach, where each cell in the cellular space is computed by a CUDA thread. A second approach concerns the adoption of a list of CA computational active cells which is handled step by step by an efficient stream compaction algorithm, in order to reduce the excessive use of computationally inactive threads. First results performed on different graphic processors have shown that, by adopting the different CUDA strategies, this kind of hardware can be effective for landslide risk mitigation.

show abstract

“…Combining MPI with CUDA has become a popular strategy for solving problems that consume more memory than is available on a single GPU [8], [9]. For such problems, the data to be shared among the MPI ranks is often stored on the GPUs.…”

Section: A Cuda-aware Mpimentioning

confidence: 99%

An Empirical Evaluation of Allgatherv on Multi-GPU Systems

Rolinger¹,

Simon²,

Krieger³

2018

2018 18th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID)

View full text Add to dashboard Cite

Applications for deep learning and big data analytics have compute and memory requirements that exceed the limits of a single GPU. However, effectively scaling out an application to multiple GPUs is challenging due to the complexities of communication between the GPUs, particularly for collective communication with irregular message sizes. In this work, we provide a performance evaluation of the Allgatherv routine on multi-GPU systems, focusing on GPU network topology and the communication library used. We present results from the OSUmicro benchmark as well as conduct a case study for sparse tensor factorization, one application that uses Allgatherv with highly irregular message sizes. We extend our existing tensor factorization tool to run on systems with different node counts and varying number of GPUs per node. We then evaluate the communication performance of our tool when using traditional MPI, CUDA-aware MVAPICH and NCCL across a suite of realworld data sets on three different systems: a 16-node cluster with one GPU per node, NVIDIA's DGX-1 with 8 GPUs and Cray's CS-Storm with 16 GPUs. Our results show that irregularity in the tensor data sets produce trends that contradict those in the OSU micro-benchmark, as well as trends that are absent from the benchmark.

show abstract

An MPI-CUDA implementation of an improved Roe method for two-layer shallow water systems

Cited by 22 publications

References 25 publications

Acceleration of a Python-Based Tsunami Modelling Application via CUDA and OpenHMPP

Acceleration of a Python-Based Tsunami Modelling Application via CUDA and OpenHMPP

CUDA Dynamic Active Thread List Strategy to Accelerate Debris Flow Simulations

An Empirical Evaluation of Allgatherv on Multi-GPU Systems

Contact Info

Product

Resources

About