A run control framework to streamline profiling, porting, and tuning simulation runs and provenance tracking of geoscientific applications

Wendy Sharples, Ilya Zhukov, Markus Geimer, Klaus Goergen, Sebastian Luehrs, Thomas Breuer, Bibi Naz, Ketan Kulkarni, Slavko Brdar, and Stefan Kollet

Download

Final revised paper (published on 13 Jul 2018)
Supplement to the final revised paper
Preprint (discussion started on 27 Oct 2017)
Supplement to the preprint

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC1: 'Some issues with scope and structure', Anonymous Referee #1, 21 Nov 2017
- AC1: 'Response to anonymous reviewer 1', Wendy Sharples, 10 Jan 2018
RC2: 'Issues with scope, uniformity and definitions', Anonymous Referee #2, 28 Nov 2017
- AC2: 'Response to anonymous reviewer 2', Wendy Sharples, 10 Jan 2018

Peer-review completion

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

AR by Wendy Sharples on behalf of the Authors (29 Jan 2018) Author's response Manuscript

ED: Referee Nomination & Report Request started (13 Feb 2018) by Steve Easterbrook

RR by Anonymous Referee #1 (29 Mar 2018)

RR by Anonymous Referee #3 (20 Apr 2018)

Suggestions for revision or reasons for rejection

This is a “second round” review of the paper following the authors’ response to the initial reviewers’ comments.

The paper raises the important issue of providing (automated) tool support for the activities of porting, profiling, tuning and running geoscience applications, and providing provenance information to enable reproducibility of runs and the validation of the science results produced by simulations as things such as compilers, computers, algorithms etc. change over time. The paper then describes a specific framework that they have developed, RCF, and illustrates its use on a specific test application of theirs (ParFLow ) running on a specific machine at their site.

The authors have done a good job of addressing the initial reviewers’ comments and this, in my opinion, has led to a much better structured and more readable paper.

I do feel the paper has a quite a narrow focus though; essentially it is a case study of one system to address some important issues. The Geoscience community is not the only one facing these problems, of course, but I think it would require a different paper to survey approaches across other areas (the revised paper has taken some steps to look elsewhere at a couple of, essentially, workflow tools but only within the context of the Geoscience community).

Below are some (minor) suggestions for improvements that the authors may find useful:

P2 11 – p3 3: the references in this section are all relatively old and a little out of date (there is nothing beyond 2014 in a rapidly developing area). It would be useful to include some more recent references. For example, GP-GPU accelerators are now well beyond “10s of cores” (p 2 25) and their performance is currently in the 10s of TFLOP/s range (rather than 1 TFLOP/s, p2 24 - e.g. see the NVIDIA Titan – and that’s a year or two old now!).
Also, a major current trend towards developing exascale machines is the exploitation of FPGAs for acceleration of both applications and communication (e.g. the recently funded EU EuroEXA project, https://euroexa.eu/, among others).

P4 9: please make it clear here that the ParFlow case study will focus on its execution on the Juqueen machine at JSC.

P6 3: change to -> “The directory structure for the RCF run harness…”

P9 12: it would be useful to more clearly link the section starting here on “health examination” to the follow up in Sec. 3.3. I suggest adding a subtitle (unnumbered) “Health Check Protocol” to make the link explicit.

P9 24/25: for completeness, define the symbols used in eqn 1. (i.e. T_1 and T_N) – other symbols in equations are defined.

P13 Fig. 3 I spotted that coll_message_size does not appear in Appendix B!

Hide

ED: Publish subject to minor revisions (review by editor) (08 May 2018) by Steve Easterbrook

AR by Wendy Sharples on behalf of the Authors (23 May 2018) Author's response Manuscript

ED: Publish as is (12 Jun 2018) by Steve Easterbrook

Short summary

Next-generation geoscientific models are based on complex model implementations and workflows. Next-generation HPC systems require new programming paradigms and code optimization. In order to meet the challenge of running complex simulations on new massively parallel HPC systems, we developed a run control framework that facilitates code portability, code profiling, and provenance tracking to reduce both the duration and the cost of code migration and development, while ensuring reproducibility.