Par@Graph – a parallel toolbox for the construction and analysis of large complex climate networks

Ihshaish, H.; Tantet, A.; Dijkzeul, J. C. M.; Dijkstra, H. A.

doi:https://doi.org/10.5194/gmd-8-3321-2015

Articles | Volume 8, issue 10

https://doi.org/10.5194/gmd-8-3321-2015

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

https://doi.org/10.5194/gmd-8-3321-2015

© 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

Par@Graph – a parallel toolbox for the construction and analysis of large complex climate networks

H. Ihshaish, A. Tantet, J. C. M. Dijkzeul, and H. A. Dijkstra

Download

Final revised paper (published on 22 Oct 2015)
Preprint (discussion started on 20 Jan 2015)

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC C36: 'Review of Ihshaish et al., Par@Graph — a parallel toolbox for the construction and analysis of large complex climate networks', Jonathan Donges, 14 Feb 2015
- AC C460: 'Authors response - Donges', Hisham Ihshaish, 15 Apr 2015
RC C112: 'Surprising performance claims require more evidence', Christian Staudt, 04 Mar 2015
- AC C466: 'Authors response', Hisham Ihshaish, 15 Apr 2015
  - AC C474: 'Runtime log file', Hisham Ihshaish, 15 Apr 2015

Peer-review completion

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

AR by Hisham Ihshaish on behalf of the Authors (16 Jun 2015) Author's response Manuscript

ED: Referee Nomination & Report Request started (24 Jun 2015) by David Ham

RR by Christian Staudt (09 Jul 2015)

Suggestions for revision or reasons for rejection

In this revision the authors have improved the presentation and addressed reviewer comments., e.g. it was clarified that they have not actually calculated betweenness in a 10^12 edge graph. This is relevant work that should be accepted subject to minor revisions.

1) The following issues should be corrected before acceptance:

- I'm not impartial on this, but the claim that NetworKit is limited to 10^7 edges is not true. Typical hardware allows for the analysis of networks with at least 10^9 edges (see [1] for reference). This probably holds for similar network analysis libraries like igraph and graph-tool. Only the 10^12 edge network is out of the range of these preexisting tools and requires distributed parallel processing. The authors do not have to rely on published figures alone: In this case, it is very easy to test the statement. Freely available, closely related software can and should be run in experiments.

- Plots in Fig. 5 are still too small to be read comfortably.

2) Here the paper could have made its points more strongly (but they are not reasons for delaying publication):

- Related to the issue above, the work could have been strengthened overall by a thorough experimental comparison to existing software tools (including comparative running time measurements). That would result in sound recommendations on when to apply a new tool like Par@graph.
Comparison to the standard igraph library would be especially relevant to show the improvements achieved by the Par@graph optimisations.

3) Lastly, some comments that are not direct criticisms of the paper but points for further discussion:

- Faced with massive network data, one solution is adding more raw computing power, another is developing more refined methods, e.g. statistical methods that give approximate but qualitatively equivalent results. Network analysis is essentially a form of statistics, and while the exact results often come at massive computational cost, a result derived from sampling could be just as good and much cheaper.

- This work applies network analysis methods to networks that result from thresholding a matrix of pairwise correlations, rather than a directly observed collection of links that are more typical in network analysis. The structure of the resulting network depends strongly on the choice of the threshold value, and there seem to be no clear guidelines on how to choose the threshold. Furthermore, it has been argued [3] that a single global threshold is inappropriate.

[1]: http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=6687351&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D6687351
[2]: http://arxiv.org/ftp/arxiv/papers/1403/1403.3005.pdf
[3]: http://www.pnas.org/content/106/16/6483.short

Hide

ED: Publish subject to minor revisions (Editor review) (10 Jul 2015) by David Ham

AR by Hisham Ihshaish on behalf of the Authors (20 Jul 2015) Author's response Manuscript

ED: Publish as is (25 Jul 2015) by David Ham

AR by Hisham Ihshaish on behalf of the Authors (03 Aug 2015) Manuscript

Short summary

Par@Graph, a software toolbox to reconstruct and analyze large-scale complex climate networks. It exposes parallelism on distributed-memory computing platforms to enable the construction of massive networks from large number of time series based on the calculation of common statistical similarity measures between them. Providing additionally parallel graph algorithms to enable fast calculation of important and common properties of the generated networks on SMP machines.