Parallel Implementation of a PETSc-Based Framework for the General Curvilinear Coastal Ocean Model
Round 1
Reviewer 1 Report
See attached.
Comments for author File: 
Comments.pdf
Author Response
05/22/2019
Dear Reviewer #1,
We want to thank you for the feedback provided, we have addressed each of the comments and feel that this new version of the paper is much improved and streamlined. I would also want to remark the immense usefulness of your previous comments to our first version of the paper from last year, which we tried following as close as possible and indeed made the manuscript much stronger.
The introduction content has been expanded and divided into more specific sections, this new introduction includes comparisons with other nonhydrostatic ocean models and all of the suggested references, we also expand on the multiscale / multiphysics GCCOM examples and how they fit in the context of coastal ocean modeling.
As requested, we have included two more seamount examples with lower resolutions than the previous one, which is still our high resolution experiment in the paper. The new examples provide a more fleshed out analysis on how problem size affects parallel scaling and efficiency, but also how a higher resolution mesh is capable of showing new a complex phenomena that is not possible to see in lower resolutions. There is a new results section dedicated to multiscale / multiphysics in the GCCOM where we have included this low resolution vs. high resolution comparison.
After some deliberation and asking the editor for his opinion, we decided not to include the lock exchange problem in this paper, the detailed reasons are commented below but at the same time we followed your suggestions as close as possible utilizing the seamount case instead.
Every change made was tracked by using red lettering so is easier to recognize from what was unchanged.
Please find annexed to this letter the full amount of reviewer comments we received with an inline resolution in blue to each one of them, so we feel we have finalized addressing them.
Kind Regards,
Manuel Valera
Author Response File: Author Response.pdf
Reviewer 2 Report
Authors have implemented a PETSc-based framework for the General Curvilinear Coastal Ocean Model, which is known to be a computationally expensive parallel model requiring large memory. As a solution for parallel optimization, the Fortran interfaced PETSc library is chosen as a framework to help reduce the complexity of managing the 3D geometry, to improve parallel algorithm design, and to provide a rich suite of choices to solve and precondition the related linear systems. A case study is presented for benchmarking.
The paper is well written, but the reviewer has a few concerns/comments, as follows.
1. Define PETSc in Abstract, like other abbreviations there.
2. On page 3, the last word ‘laplacian’ should be capitalized.
3. All equation, figure, and table numbers in the text are written with small letter. For example, the first equation is referred to as ‘equation 1’. Since a specific equation is referred to, it should be written as ‘Equation 1’. All instances (e.g., equations, figures, and tables) should be corrected.
4. Equations 3-5 seem to have multiple equations under each number. This is unconventional. Each equation should have one number. Authors may consider (3a), (3b), etc., if needed
5. Table 2 title is too long. Explanation should be in the text body, not in the title.
6. It requires some time to figure out what Equation 14 is. I wonder defining the generalization defeats the purpose of simplification.
7. On page 9, the matrix A indices are separated with asterisks (*). Perhaps comma or multiplication sign would be better.
8. Only one case is presented. Any practical problem for which this model is written for can be presented? No need to try all different number of processors. Perhaps results from one or two choices of processors compared against those from serial case would be good enough.
9. The first paragraph of subsection 3.1.3 on page 12 should be revised. The sentence starting with ‘Also’ does not make much sense.
10. The Abstract claims that PETSc is implemented to take the advantage of a suite of solvers and preconditioners. Some form of validation should be given to show the impact of solver and preconditioner choices. Again, results from one or two choices of processors compared against those from serial case would be good enough.
Author Response
05/22/2019
Dear Reviewer #2,
We want to thank you for the feedback provided, we have addressed each of the comments and feel that this new version of the paper is much improved and streamlined.
The introduction content has been expanded and divided into more specific sections, this new introduction includes comparisons with other nonhydrostatic ocean models and all of the suggested references, we also expand on the multiscale / multiphysics GCCOM examples and how they fit in the context of coastal ocean modeling.
As requested, we have included two more seamount examples with lower resolutions than the previous one, which is still our high resolution experiment in the paper. The new examples provide a more fleshed out analysis on how problem size affects parallel scaling and efficiency, but also how a higher resolution mesh is capable of showing new a complex phenomena that is not possible to see in lower resolutions. There is a new results section dedicated to multiscale / multiphysics in the GCCOM where we have included this low resolution vs. high resolution comparison.
Every change made was tracked by using red lettering so is easier to recognize from what was unchanged.
Please find annexed to this letter the full amount of reviewer comments we received with an inline resolution in blue to each one of them, so we feel we have finalized addressing them.
Kind Regards,
Manuel Valera
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Manuscript is significantly improved. Thank you for your work throughout. Introduction is informative. Moving tables to appendix is fine. Additional detail on seamount is good.
State variable name in fig 9 and 10 captions. White space in fig 9 does not make sense - is it a 3D view? Use same colorbar extents in 9 and 10 between left and right, at least on top panels, so we can see the sims do the same thing at high and low res. Don't use contour lines if they fill in the plot, like fig 10 top left. Contour lines near zero don't tell us anything.
Author Response
Thanks for your time reviewing our paper and providing important feedback.
The figures have been updated to reflect the same color scale in the top panels. The scales and similarities have been preserved whenever possible.
On the white space: This is an effect of the curvilinear grid, since it is a curved region of the grid we are plotting this region will be distorted in some way, hence the white space. Notice the corners of the region touch each side of the plot, this is a way to visualize the curvilinear nature of the grid.
The contour levels have been tweaked to minimize the overlap, creating a clearer image.
Once again thanks,
Manuel Valera
