A parallelization scheme to simulate reactive transport in the  subsurface environment with OGS#IPhreeqc 5.5.7-3.1.2

He, W.; Beyer, C.; Fleckenstein, J. H.; Jang, E.; Kolditz, O.; Naumov, D.; Kalbacher, T.

doi:https://doi.org/10.5194/gmd-8-3333-2015

Articles | Volume 8, issue 10

https://doi.org/10.5194/gmd-8-3333-2015

© Author(s) 2015. This work is distributed under
the Creative Commons Attribution 3.0 License.

https://doi.org/10.5194/gmd-8-3333-2015

© Author(s) 2015. This work is distributed under
the Creative Commons Attribution 3.0 License.

Articles | Volume 8, issue 10

Development and technical paper

|

22 Oct 2015

Development and technical paper |

| 22 Oct 2015

A parallelization scheme to simulate reactive transport in the subsurface environment with OGS#IPhreeqc 5.5.7-3.1.2

W. He, C. Beyer, J. H. Fleckenstein, E. Jang, O. Kolditz, D. Naumov, and T. Kalbacher

Download

Final revised paper (published on 22 Oct 2015)
Supplement to the final revised paper
Preprint (discussion started on 03 Mar 2015)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC C273: 'report and comments', Anonymous Referee #1, 26 Mar 2015
- AC C478: 'Interactive comment on “A parallelization scheme to simulate reactive transport in the subsurface environment with OGS#IPhreeqc” by W. He et al.', Wenkui He, 18 Apr 2015
SC C979: 'Executive Editor Comment on "A parallelization scheme to simulate reactive transport in the subsurface environment with OGS#IPhreeqc"', Astrid Kerkweg, 05 Jun 2015
- SC C1075: 'reply to Executive editor: Code availability', Wenkui He, 10 Jun 2015
RC C1166: 'Reviewer comments on OGS#Iphreeqc coupling parallel performance', Glenn Hammond, 18 Jun 2015
- AC C1627: 'Response to Review from Glenn Hammond', Wenkui He, 07 Aug 2015

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

AR by Wenkui He on behalf of the Authors (07 Aug 2015) Author's response Manuscript

ED: Referee Nomination & Report Request started (15 Aug 2015) by Hilary McMillan

RR by Glenn Hammond (15 Sep 2015)

Suggestions for revision or reasons for rejection

Thank you for your responses to the previous questions. I am satisfied with the responses. I apologize for the additional questions below, but there is so much new content in the revised paper that I have additional questions/concerns.

The grammar in the paper is awkward in many locations. I highly recommend that a native English speaker edit/revise some of the wording. Several situations are listed below, but there are many more. This should not be too difficult.

(Page) 2: Line 4-5: “…and more straightforwardly on high performance computers” – needs to be revised as this does not make sense. Easier?

2: Line 10: “almost any chemical reaction”? How about “many chemical reactions” or “numerous chemical reactions”?

3:Line 8: “coupling of different codes…is an indispensable choice” – this is solely opinion.

4: Line 7: Since Hammond et al., 2012 is cited but not included in the references, you may want to review all citations to ensure that the references exist at the end.

5:Line 4: “therefore a speedup for flow/transport is not given anymore”. Should this be “is not experienced anymore”.

5:Line 12: “flexible distribution of different amount of computer resources” -> amounts?

6:Line 25: “It provides a well-defined series of methods” How about “it provides an API (Application Programming Interface)”.

7: Line 18: The reference to CMake is out of the blue. How about calling it the build system? Otherwise, you need to explain CMake.

9: Line 19: Lasaga type rate law -> it is referred to as “transition state theory” (and then cite Lasaga)

10: Line 5: “IPhreeqc interface and the overhead involved in calling IPhreeqc are 12.7% and 3.8%”. Can you point me to where you explain/define the difference between the interface and overhead? I assume that the initialization of the thermodynamic database is in the overhead….

11: Line 12: The description of domain decomposition leads me to believe that you are using Richardson iteration (i.e. where one solves the local subdomain and then updates the ghost/halo values), but I find that hard to believe. With domain decomposition, the linear/nonlinear system is solved across the entire domain for each iteration. If you are truly iterating between subdomains, I highly recommend using the full implicit solve as it would converge much faster.

11: Line 24: Again, this doesn’t sound like traditional domain decomposition.

12: Line 21: “a single input string will be prepared” – is there a separate string for each cell (all concentrations in each cell) or is there a single string for all concentrations in all cells being solved on a core?

13: Line 11: Are the concentrations in the strings full (double) precision?

13: Line 28: An MPI_Allreduce() creates a copy of the full buffer for each process. So, you are generating a full global concentration vector for each core? Are you sure that this is not a gather or scatter operation? That seems more reasonable.

15: Line 21: To me it seems that from Figure 8b, the best (fastest) combination would be 18 subdomains (best flow and transport speedup) + 20 geochemistry cores (best chemistry speedup). Just a comment.

15: Line 28: My experience is that parallel efficiency can degrade significantly well before the number of border (ghost) nodes approaches the total number of nodes.

16: Line 14: Averagely -> On average?

16: Line 16: Let the reader come to this conclusion on their own as some may disagree.

18: Line 9: The addition of the uranium leaching problem, an example that clearly illustrates the benefit of this algorithm, greatly improves the paper.

20: Line 14: I believe that where this new algorithm will pay off is with massively parallel computation (hundreds of thousands of cores) where linear solver performance is poor. In those cases, one would want to use fewer cores to solve flow and transport, as conventional parallel preconditioners tend to degrade in performance at extremely large core counts.

Hide

ED: Publish subject to minor revisions (Editor review) (16 Sep 2015) by Hilary McMillan

AR by Wenkui He on behalf of the Authors (25 Sep 2015) Author's response Manuscript

ED: Publish as is (07 Oct 2015) by Hilary McMillan

Short summary

This technical paper presents a new tool to simulate reactive transport processes in subsurface systems and which couples the open-source software packages OpenGeoSys and IPhreeqc. A flexible parallelization scheme was developed and implemented to enable an optimized allocation of computer resources. The performance tests of the coupling interface and parallelization scheme illustrate the promising efficiency of this generally valid approach to simulate reactive transport problems.