Journal metrics

Journal metrics

  • IF value: 4.252 IF 4.252
  • IF 5-year value: 4.890 IF 5-year 4.890
  • CiteScore value: 4.49 CiteScore 4.49
  • SNIP value: 1.539 SNIP 1.539
  • SJR value: 2.404 SJR 2.404
  • IPP value: 4.28 IPP 4.28
  • h5-index value: 40 h5-index 40
  • Scimago H index value: 51 Scimago H index 51
Discussion papers | Copyright
https://doi.org/10.5194/gmd-2017-301
© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 4.0 License.

Methods for assessment of models 22 Jan 2018

Methods for assessment of models | 22 Jan 2018

Review status
This discussion paper is a preprint. A revision of this manuscript was accepted for the journal Geoscientific Model Development (GMD) and is expected to appear here in due course.

Simulation of the Performance and Scalability of MPI Communications of Atmospheric Models running on Exascale Supercomputers

Yongjun Zheng and Philippe Marguinaud Yongjun Zheng and Philippe Marguinaud
  • Centre National de Recherches Météorologiques, Météo France, Toulouse 31057, France

Abstract. In this study, we identify the key MPI operations required in atmospheric modelling; then, we use a skeleton program and a simulation framework (based on SST/macro simulation package) to simulate these MPI operations (transposition, halo exchange, and allreduce), with the perspective of future exascale machines in mind. The experimental results show that the choice of the collective algorithm has a great impact on the performance of communications, in particular we find that the generalized ring-k algorithm for the alltoallv operation and the generalized recursive-k algorithm for the allreduce operation perform the best. In addition, we observe that the impacts of interconnect topologies and routing algorithms on the performance and scalability of transpositions, halo exchange, and allreduce operations are significant, however, that the routing algorithm has a negligible impact on the performance of allreduce operations because of its small message size. It is impossible to infinitely grow bandwidth and reduce latency due to hardware limitations, thus, congestion may occur and limit the continuous improvement of the performance of communications. The experiments show that the performance of communications can be improved when congestion is mitigated by a proper configuration of the topology and routing algorithm, which uniformly distribute the congestion over the interconnect network to avoid the hotspots and bottlenecks caused by congestion. It is generally believed that the transpositions seriously limit the scalability of the spectral models. The experiments show that although the communication time of the transposition is larger than those of the wide halo exchange for the Semi-Lagrangian method and the allreduce in the GCR iterative solver for the Semi-Implicit method below 200000 MPI processes, the transposition whose communication time decreases quickly as the number of MPI processes increases demonstrates strong scalability in the case of very large grids and moderate latencies; the halo exchange whose communication time decreases more slowly than that of transposition as the number of MPI processes increases reveals its weak scalability; in contrast, the allreduce whose communication time increases as the number of MPI processes increases does not scale well. From this point of view, the scalability of the spectral models could still be acceptable, therefore it seems to be premature to conclude that the scalability of the grid-point models is better than that of spectral models at exascale, unless innovative methods are exploited to mitigate the problem of the scalability presented in the grid-point models.

Download & links
Yongjun Zheng and Philippe Marguinaud
Interactive discussion
Status: closed
Status: closed
AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment
Printer-friendly Version - Printer-friendly version Supplement - Supplement
Interactive discussion
Status: closed
Status: closed
AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment
Printer-friendly Version - Printer-friendly version Supplement - Supplement
Yongjun Zheng and Philippe Marguinaud
Yongjun Zheng and Philippe Marguinaud
Viewed
Total article views: 634 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
451 156 27 634 12 29
  • HTML: 451
  • PDF: 156
  • XML: 27
  • Total: 634
  • BibTeX: 12
  • EndNote: 29
Views and downloads (calculated since 22 Jan 2018)
Cumulative views and downloads (calculated since 22 Jan 2018)
Viewed (geographical distribution)
Total article views: 625 (including HTML, PDF, and XML) Thereof 624 with geography defined and 1 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Cited
Saved
Discussed
No discussed metrics found.
Latest update: 18 Aug 2018
Publications Copernicus
Download
Short summary
The impacts of collective algorithms, interconnect topologies, and routing algorithms on the performance and scalability of transpositions, halo exchange, and allreduce operations are significant. The performance of communications can be improved when the congestion is mitigated by a proper configuration of the topology and routing algorithm. The scalability of the spectral models could be acceptable for exascale supercomputers.
The impacts of collective algorithms, interconnect topologies, and routing algorithms on the...
Citation
Share