A distributed computing approach to improve the performance of the Parallel Ocean Program (v2.1)

Werkhoven, Ben van; Maassen, Jason; Kliphuis, Michael; Dijkstra, Henk A.; Brunnabend, Sandra-Esther; Meersbergen, Maarten van; Seinstra, F.J.; Bal, Henri E.

doi:10.5194/gmd-7-267-2014

Cited by 13 publications

(7 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The algorithm takes the geographical features of the planet into account, in particular, the fact that communication with land blocks is not needed. Our approach is described extensively in [15] and was shown to outperform even the available strategies in traditional (non-distributed) runs.…”

Section: Technical and Non-technical Complexitiesmentioning

confidence: 99%

“…However, no exchange is needed with neighboring blocks that only contain land. Hence, by taking geography into account, the reduced communication between specific blocks can be exploited to distribute the model over multiple supercomputers [15]. Improve ocean model performance using hardware accelerators.…”

Section: Execution Scenariosmentioning

confidence: 99%

“…Apart from the lack of real hotspots in POP, the main challenge is to overcome the PCIe bus bottleneck, by maximizing the overlap of CPU-GPU data transfers with computation. Our solutions to this problem are described in [15,16,34], and our GPU-enabled POP implementation is made available under a free open source license [35]. This is a stand-alone version of the ocean model, however, not integrated in CESM.…”

Section: Technical and Non-technical Complexitiesmentioning

confidence: 99%

“…We have made good progress in the scaling up of the ocean modeling itself [14], and in the implementation and optimization of the distributed and hierarchical CESM simulations [15], with our work on GPU optimization receiving a best paper award nomination at CCGrid 2014 [16]. We have also made some progress in the distributed execution over a subset of the supercomputing resources available to us, the results of which are currently prepared for a separate publication.…”

mentioning

confidence: 99%

See 3 more Smart Citations

On the complexities of utilizing large‐scale lightpath‐connected distributed cyberinfrastructure

Maassen

Werkhoven

Meersbergen

et al. 2016

Concurrency and Computation

View full text Add to dashboard Cite

EDITORIAL Report on the first workshop on negative and null results in eScience 1. INTRODUCTION New techniques and technologies, such as the use of large-scale computing, influence research approaches, methods, and scales and are rapidly changing the scientific landscape. Research projects in eScience thus start with many assumptions and many unknowns and are often complex. While the scientific process is sometimes viewed, at least in hindsight, as a linear progression from one good idea to the next, it is in fact fraught with false starts, wrong assumptions, and dead ends. The increasing reliance on computation adds to the scope of problems that occur. Researchers invest a significant amount of time and effort in their research. Funding agencies similarly make large investments to support such research, on the assumption that most of the research will be successful. When the research assumptions and hypotheses turn out to be false, causing results that are "negative" or "null", the natural bias is to judge that the research project "failed." The history of science, however, shows that negative results may be an opportunity to revolutionize a field of study. For example, Fleming noticed that his flu cultures were contaminated by mold, but that there was infection around that mold, leading to his discovery of Penicillin. Similarly, a project today may fail because of the misuse or failure of computational support. Such "failures" actually indicate that there is an opportunity for the cyberinfrastructure research community to improve computing resources and tools. The interaction of these modes of failure is multi-faceted. Negative results have been difficult to find in published papers in all scientific domains. We identify three reasons for this. First, negative results may not be identified as such but simply considered mistakes. Such cases may never be investigated further. Secondly, paper referees may demand a higher standard from such results, because they are more difficult to understand or challenge the conventional narrative. Third, researchers may self-select against publishing such results in light of the previous point. Our experience with and observations of numerous projects in the area of eScience inspired us to create a workshop as a forum to initiate the discussion of negative and null results in the context of eScience (Section 4). Our effort attempts to systematically approach problems voiced in the popular media [1, 2] about the importance of negative results. This paper encapsulates the topics presented and lessons learned at this workshop, including efforts to: (1) Examine the topics surrounding the notion of negative and null results. Classify and propose a taxonomy of negative and null results in eScience. (2) Understand the causes of negative results. Discuss the activities and actions undertaken in order to mitigate the effects of negative results. (3) Challenge confirmation bias and the prejudice against negative results. Consider how to present negative results to the scientific communi...

show abstract

Section: Technical and Non-technical Complexitiesmentioning

confidence: 99%

Section: Execution Scenariosmentioning

confidence: 99%

Section: Technical and Non-technical Complexitiesmentioning

confidence: 99%

mentioning

confidence: 99%

See 2 more Smart Citations

On the complexities of utilizing large‐scale lightpath‐connected distributed cyberinfrastructure

Maassen

Werkhoven

Meersbergen

et al. 2016

Concurrency and Computation

View full text Add to dashboard Cite

show abstract

“…In the eSalsa project, we enabled global ocean modeling at very high spatial resolution using GPU computing technologies in collaboration with Utrecht University. An existing numerical ocean model code was optimized for execution on the CPU-GPU Cartesius supercomputer of SURF-Sara [13]. Century scale simulations, up to 2100 under climate change, are extremely challenging at the eddy-resolving spatial resolution (equivalent to about 10 km horizontal scale; [4]).…”

Section: Fig 4 Screenshot Of the Research Software Directorymentioning

confidence: 99%

eScience development and experiences in The Netherlands

Hazeleger

Bakker

Nieuwpoort

2018

Informatik Spektrum

View full text Add to dashboard Cite

The Netherlands eScience Center is the national expertise center for the development and application of research software. Collaborating with researchers from all academic disciplines, it extends the breadth and depth of research opportunities by exploiting the latest insights from computer and data science, as well as making optimal use of hardware, software, and data infrastructures. It does so through problem-driven research projects where eScience research engineers, employed by the eScience Center, collaborate with researchers in all disciplines at Dutch academic institutions. Project software is generalized and made available for reuse for other disciplines and goals. The center has three main technological competences: efficient computing, optimized data handling, and data analytics. Furthermore, on the national level it coordinates and contributes to science policies on computing, data, and applications thereof. With its two main assets, a staff of highly educated and multi-disciplinary eScience Research Engineers and an open online directory of research software tools and knowledge, it successfully contributes to the Dutch scientific landscape and enhances and accelerates all research in The Netherlands and beyond.

show abstract