More than one hundred days were simulated over very large domains with fine (0.156 km to 2.5 km) grid spacing for realistic conditions to test the hypothesis that storm (kilometer) and large-eddy (hectometer) resolving simulations would provide an improved representation of clouds and precipitation in atmospheric simulations. At scales that resolve convective storms (storm-resolving for short) scales, the vertical velocity variance becomes resolved and a better physical basis is achieved for representing clouds and precipitation. Similar to past studies we find an improved representation of precipitation at kilometer scales, as compared to models with parameterised convection. The main precipitation features (location, diurnal cycle and spatial propagation) are well captured already at kilometer scales, and refining resolution to hectometer scales does not substantially change the simulations in these respects. It does, however, lead to a reduction in the precipitation on the timescales considered-most notably over the Tropical ocean. Changes in the distribution of precipitation, with less frequent extremes are also found in simulations incorporating hecto-meter scales. Hectometer scales appear more important for the representation of clouds, and make it possible to capture many important aspects of the cloud field, from the vertical distribution of cloud cover, to the distribution of cloud sizes, to the diel (daily) cycle. Qualitative improvements, particularly in the ability to differentiate cumulus from stratiform clouds, are seen when reducing the grid spacing from kilometer to hectometer scales. At the hectometer scale new challenges arise, but the similarity of observed and simulated scales, and the more direct 1 connection between the circulation and the unconstrained degrees of freedom make these challenges less daunting. This quality, combined with an already improved simulation as compared to more parameterised models, underpins our conviction that the use and further development of storm-resolving models offers exciting opportunities for advancing understanding of climate and climate change.

We discuss scientific features and computational performance of kilometre-scale global weather and climate simulations, considering the Icosahedral Non-hydrostatic (ICON) model and the Integrated Forecast System (IFS). Scalability measurements and a performance modelling approach are used to derive performance estimates for these models on upcoming exascale supercomputers. This is complemented by preliminary analyses of the model data that illustrate the importance of high-resolution models to gain improvements in the accuracy of convective processes, a better understanding of physics dynamics interactions and poorly resolved or parametrized processes, such as gravity waves, convection and boundary layer. This article is part of the theme issue ‘Multiscale modelling, simulation and computing: from the desktop to the exascale’.

Abstract. State-of-the-art Earth system models typically employ grid spacings of O(100 km), which is too coarse to explicitly resolve main drivers of the flow of energy and matter across the Earth system. In this paper, we present the new ICON-Sapphire model configuration, which targets a representation of the components of the Earth system and their interactions with a grid spacing of 10 km and finer. Through the use of selected simulation examples, we demonstrate that ICON-Sapphire can (i) be run coupled globally on seasonal timescales with a grid spacing of 5 km, on monthly timescales with a grid spacing of 2.5 km, and on daily timescales with a grid spacing of 1.25 km; (ii) resolve large eddies in the atmosphere using hectometer grid spacings on limited-area domains in atmosphere-only simulations; (iii) resolve submesoscale ocean eddies by using a global uniform grid of 1.25 km or a telescoping grid with the finest grid spacing at 530 m, the latter coupled to a uniform atmosphere; and (iv) simulate biogeochemistry in an ocean-only simulation integrated for 4 years at 10 km. Comparison of basic features of the climate system to observations reveals no obvious pitfalls, even though some observed aspects remain difficult to capture. The throughput of the coupled 5 km global simulation is 126 simulated days per day employing 21 % of the latest machine of the German Climate Computing Center. Extrapolating from these results, multi-decadal global simulations including interactive carbon are now possible, and short global simulations resolving large eddies in the atmosphere and submesoscale eddies in the ocean are within reach.

Abstract. Classical numerical models for the global atmosphere, as used for numerical weather forecasting or climate research, have been developed for conventional central processing unit (CPU) architectures. This now hinders the employment of such models on current top performing supercomputers, which achieve their computing power with hybrid architectures, mostly using graphics processing units (GPUs). Thus also scientific applications of such models are restricted to the lesser computer power of CPUs. Here we present the development of a GPU enabled version of the ICON atmosphere model (ICON-A) motivated by a research project on the quasi-biennial oscillation (QBO), a global scale wind oscillation in the equatorial stratosphere that depends on a broad spectrum of atmospheric waves, which origins from tropical deep convection. Resolving the relevant scales, from a few km to the size of the globe, is a formidable computational problem, which can only be realized now on top performing supercomputers. This motivated porting ICON-A, in the specific configuration needed for the research project, in a first step to the GPU architecture of the Piz Daint computer at the Swiss National Supercomputing Centre, and in a second step to the Juwels-Booster computer at the Forschungszentrum Jülich. On Piz Daint the ported code achieves a single node GPU vs. CPU speed-up factor of 6.3, and now allows global experiments at a horizontal resolution of 5 km on 1024 computing nodes with 1 GPU per node with a turnover of 48 simulated days per day. On Juwels-Booster the more modern hardware in combination with an upgraded code base allows for simulations at the same resolution on 128 computing nodes with 4 GPUs per node and a turnover of 133 simulated days per day. Additionally, the code still remains functional on CPUs as it is demonstrated by additional experiments on the Levante compute system at the German Climate Computing Center. While the application shows good weak scaling making also higher resolved global simulations possible, the strong scaling on GPUs is relatively weak, which limits the options to increase turnover with more nodes. Initial experiments demonstrate that the ICON-A model can simulate downward propagating QBO jets, which are driven by wave meanflow interaction.

Classical numerical models for the global atmosphere, as used for numerical weather forecasting or climate research, have been developed for conventional central processing unit (CPU) architectures. This hinders the employment of such models on current top-performing supercomputers, which achieve their computing power with hybrid architectures, mostly using graphics processing units (GPUs). Thus also scientific applications of such models are restricted to the lesser computer power of CPUs. Here we present the development of a GPU-enabled version of the ICON atmosphere model (ICON-A), motivated by a research project on the quasi-biennial oscillation (QBO), a globalscale wind oscillation in the equatorial stratosphere that depends on a broad spectrum of atmospheric waves, which originates from tropical deep convection. Resolving the relevant scales, from a few kilometers to the size of the globe, is a formidable computational problem, which can only be realized now on top-performing supercomputers. This motivated porting ICON-A, in the specific configuration needed for the research project, in a first step to the GPU architecture of the Piz Daint computer at the Swiss National Supercomputing Centre and in a second step to the JUWELS Booster computer at the Forschungszentrum Jülich. On Piz Daint, the ported code achieves a single-node GPU vs. CPU speedup factor of 6.4 and allows for global experiments at a horizontal resolution of 5 km on 1024 computing nodes with 1 GPU per node with a turnover of 48 simulated days per day. On JUWELS Booster, the more modern hardware in combination with an upgraded code base allows for simulations at the same resolution on 128 computing nodes with 4 GPUs per node and a turnover of 133 simulated days per day. Additionally, the code still remains functional on CPUs, as is demonstrated by additional experiments on the Levante compute system at the German Climate Computing Center. While the application shows good weak scaling over the tested 16fold increase in grid size and node count, making also higher resolved global simulations possible, the strong scaling on GPUs is relatively poor, which limits the options to increase Published by Copernicus Publications on behalf of the European Geosciences Union. 6986 M. A. Giorgetta et al.: The ICON-A model on GPUs turnover with more nodes. Initial experiments demonstrate that the ICON-A model can simulate downward-propagating QBO jets, which are driven by wave-mean flow interaction.

Abstract. State-of-the-art Earth System models typically employ grid spacings of O(100 km), too coarse to explicitly resolve main drivers of the flow of energy and matter across the Earth System. In this paper, we present the new ICON-Sapphire model configuration, which targets a representation of the components of the Earth System and their interactions with a grid spacing of 10 km and finer. Through the use of selected simulation examples, we demonstrate that ICON-Sapphire can already now (i) be run coupled globally on seasonal time scales with a grid spacing of 5 km and on monthly time scales with a grid spacing of 2.5 km, (ii) resolve large eddies in the atmosphere using hectometer grid spacings on limited-area domains in atmosphere-only simulations, (iii) resolve submesoscale ocean eddies by using a global uniform grid of 1.25 km or a telescoping grid with a finest grid spacing of 530 m, the latter coupled to a uniform atmosphere and (iv) simulate biogeochemistry in an ocean-only simulation integrated for 4 years at 10 km. Comparison to observations of these various configurations reveals no obvious pitfall. The throughput of the coupled 5-km global simulation is 126 simulated days per day employing 21 % of the latest machine of the German Climate Computing Center. Extrapolating from these results, multi-decadal global simulations including interactive carbon are now possible and short global simulations resolving large eddies in the atmosphere and submesoscale eddies in the ocean are within reach.

Δημοσιεύσεις μελών--ΣΕΦ--Τμήμα Μηχανικών Πληροφορικής, 1996Parallel genetic algorithms (PGA) offer a natural and productive way to solve a problem better than a single population. Up to now the different PGA paradigms use the same evolution behaviour on each population. This paper proposes a method, called Co-operating Populations with Different Evolution Behaviours (CoPDEB) where the populations are allowed to exhibit different evolution behaviours. This is achieved by using a variety of selection mechanisms, operators, communication methods and parameters as it is explained in the sequel. This method has been tested on the problem of training a recurrent artificial neural network (RANN

This paper presents the partial results of an extended replicated pair programming experiment, conducted at the Alexander Technological Educational Institute of Thessaloniki, Greece during the two semesters of the year 2009-10. The entire study involved 160 second semester students attending a Software Engineering advanced programming course with pair programming and test-driven development as teaching methodology. A series of formal experiments were conducted to investigate, as in the first experiment, the impact of developer personalities and temperaments on pair performance. In this paper we report partial results and we focus on pair performance in terms of communication, time to complete assignments and overall score. The main variables were statistically analyzed in order to test differences between two personality groups: A control group with homogeneous personalities in pairs and an experimental one with heterogeneous personalities. The findings reported here concern only the first winter semester and confirm that pairs consisting of heterogeneous personalities and temperaments performed better than pairs with the same personality and temperament type. However, time to complete assignments was not statistically different between the two groups.