Cluster computers are dominating high performance computing (HPC) today. Basically, such machines for massively parallel processing (MPP) are set up from commodity building blocks. The success of this architecture is based on the fact that it profits from the improvements provided by mainstream computing well known under the label of Moore's law.

With systems at Petascale (10^15 operations per second) in production today the next goal in HPC is to reach Exascale (10^18 operations per second) by the end of the decade. Obviously, this target introduces new challenges. First of all there are technological problems like energy efficiency or resiliency to be overcome. Furthermore, it is questionable, if general purpose CPUs will still be competitive from an energy efficiency point of view with more specialized solutions like accelerators, namely GPUs. The scalability of today's systems is limited by the way accelerators are employed. Therefore it will become a necessity to review the idea of the cluster architecture in HPC in order to prolong their success into the future.

In order to find possible directions for next generation supercomputers we review Amdahl's and Gustafson's thoughts on scalability. Based on this analysis we propose an advance architecture combining a Cluster with a so called Booster element comprising of accelerators interconnected by a high performance fabric. We argue that this architecture provides significant advantages compared to today's accelerated clusters and might pave the way for clusters into the era of Exascale computing.

The EU-project DEEP is aiming for an implementation of this concept. This includes both, the actual hardware of a Booster system based on Intel's XEON Phi processor-architecture and the EXTOLL high-performance interconnect, and an advanced software-stack required to operate and use the Booster hardware. Besides the actual system-level layers of software the latter includes forward-looking programming paradigms that shall enable application-programmers to express the various levels of scalability embedded in their problems in a straight-forward and maintainable way. It comprehends runtime-environments optimizing the use of the proposed hardware-architecture.

Along this line six applications from fields having the potential to exploit Exascale systems will be ported to DEEP enabling for a co-development in the fields of programming-models and runtime-systems for HPC. We analyze one application from the field of space-weather in detail and explore the consequences of the constraints of the DEEP systems on its scalability.

The simulation of large portions of the heliosphere (e.g., the magnetosphere) with a kinetic description is still an unattainable goal both with the current petascale computing and with the upcoming exascale computing. This has two causes: not only the wide gap in orders of magnitude between the spatial scales of interest in the system and the size of the system itself, but also the stability constraints of kinetic methods, which impose to resolve spatial scales smaller than the scales of interest for purely numerical reasons, thus increasing the computational requirements of the simulation.
This second aspect can be made less critical with a little shrewdness. Implicit algorithms can be used instead of explicit ones to relax the stability constraints, thus allowing to use bigger grid spacings. Instead of simulating the entire system with the same grid spacing, different parts of the domain may be simulated with different resolutions chosen in order to resolve the local scales of interest rather than the smallest one in the entire system.
The Multi Level Multi Domain (MLMD) method we present here has been designed following the above mentioned lines. The full domain is simulated as a collection of an arbitrary number of levels simulated fully in fields and particles with increasing grid resolution. This way, the expensive high resolutions are used only when needed, rather than imposing a very small grid spacing also in portions of the grid where it is not required by the physics of interest. Moreover, an implicit moment method (IMM) is used to advance fields and particles on each level to substitute the strict stability constraints of explicit methods with the less strict accuracy constraint of the IMM. Such a choice also grants increased freedom in choosing the refinement factor (RF) to be used between the levels also when the time spacing is kept fixed in the system.
In this work, the major differences between MLMD and Adaptive Mesh Refinement (AMR) algorithms, mostly concerning the treatment of particles at the refinement levels, will be addressed. Communication and interlocking operations between the grids (boundary condition interpolation from the coarse to the refined grids, field projection form the refined to the coarse grids and particle repopulation at the boundary of the refined grids) will be explained in details and, finally, test simulations exploring the field and particle structures evolution across the interlocked grids in 1D and 2D will be presented, together with performance considerations.

We present results of a 3D global magnetohydrodynamic simulation of the solar wind interaction with the Earth's magnetosphere driven by the time-varying NASA Advanced Composition Explorer (ACE) satellite data during the April 6th, 2000 event. It is shown that the upstream solar wind plasma parameters enter the low-beta switch-on regime for several instants during a magnetic storm causing a complex dimpled bow shock structure. We also investigate the trace of such bow shock structures in the steady state results for certain parameter values during the event, when the solar wind plasma is in the low-beta switch-on regime, as well as during time-dependent simulations of the event. We utilize a 3D, implicit, parallel, unstructured grid, compressible finite volume ideal MHD solver for our simulations. This solver is implemented inside COOLFluiD, which is an object-oriented multi-physics framework into which we plan to implement different space weather models and couple them with the abovementioned global magnetohydrodynamic model in the future.

The Open Geospace General Circulation Model (OpenGGCM) is a coupled model of the outer magnetosphere, the magnetosphere - ionosphere (MI) coupling region, the inner magnetosphere and the ring current, the ionosphere, and the thermosphere. The outer magnetosphere part solves the time-dependent MHD equations or Hall-MHD equations. The inner magnetosphere sub model is either the Rice Convection Model (RCM) or the Comprehensive Ring Current Model (CRCM). MI coupling is primarily based on empirical models that compute electron precipitation parameters. The latter, along with the ionosphere potential, are fed into the Coupled Thermosphere Ionosphere Model (CTIM), which is by itself a fully dynamical 3d model of thermospheric and ionospheric species and their interactions. CTIM in turn provides conductance and dynamo currents back to the ionosphere potential solver.

The OpenGGCM has been in development for some 20y and has been used for numerous studies. We shall present examples relevant to space weather, such as neutral upwelling events relevant to satellite drag, ground magnetic perturbations relevant to ground induced currents, and event studies of substorms and storms. We will also discuss some recent developments, such as the use of advanced computer architectures, Hall-MHD, and data assimilation via an Ensemble Kalman Filter.

SWIFF (Space Weather Integrated Forecasting Framework) is a FP7 project that has for objective to develop an integrated framework for the physics modelling of space weather from the solar corona to the Earth.

Concerning the coupling at the Earth, a 3D dynamic model of the plasmasphere has been developed and coupled to the ionospheric IRI model. The three dimensional coupled model determines at any chosen time the number density and the temperatures of the electrons and ions for altitudes from 60 km to the position of the plasmapause and even at higher radial distances in the plasmaspheric trough. The plasmapause position is highly variable depending on the geomagnetic activity.

The polar wind is also modeled with a similar kinetic approach as used for the plasmasphere, but considering open magnetic field lines at high latitudes. The model determines the profiles of all the moments of the particles, and especially density, temperatures, escape flux and heat flux.

Finally, the outer electron belt is highly variable with space weather. During geomagnetic storms, the electron fluxes vary from several orders of magnitudes and the outer belt penetrates closer to the Earth. Links between the dynamics of the radiation belts and the position of the plasmapause have been identified with satellite observations.

The links between these different regions of the inner magnetosphere and the principles of the different dynamic models based on the kinetic approach that are developed at BISA will be described. The results of these models are made available to the scientific community by free run on the space weather portal (www.spaceweather.eu).

A crucial objective of space weather modelling is forecasting the arrival of Solar Energetic
Particles (SEPs) and their intensities at a given location in space. The FP7 COMESEP (Coronal Mass Ejections and Solar Energetic Particles: Forecasting the Space Weather Impact) project includes modelling of both SEP and CME propagation. The SEP propagation model is a full-orbit test particle numerical code that allows transport across the mean magnetic field to be taken into account and flexibility in the definition of the large-scale interplanetary magnetic field configuration. Time dependent injection functions are studied to simulate particle acceleration due to a CME shock front and the resulting SEP intensities measured e.g. near Earth investigated. Within COMESEP, the objective is to link the detection and propagation modelling of CMEs with the SEP model, where information about the CME's characteristics is an input to the test particle simulations. We present initial results from this modelling effort. This work has received funding from the European Commission FP7 Project COMESEP (263252).

A growing body of theoretical and observational evidence suggests that solar energetic particles may gain most of their energy at traveling shocks relatively close to the Sun. The observed and modeled Alfven speed profiles in the corona allow for fast shocks to easily develop within 20 solar radii. By coupling global MHD simulation results for a case study coronal mass ejection and related shock with a global energetic particle acceleration and transport kinetic model, we investigate SEP acceleration in the three dimensional corona. We show that the shock and various plasma structures may efficiently accelerate suprathermal protons to hundreds of MeV energies during their coronal transit. The resulting SEP flux spectra vary greatly depending on the latitudes and longitudes of the guiding field lines. We follow the proton fluxes to 1 AU and determine their longitudinal dependence. This result may confirm shocks as the dominant mechanism for creation of energetic particles in the vicinity of the Sun, and help explain the variation in observed SEP time series at 1 AU.

The EU/FP7 project SEPServer will develop simulation tools for modeling the acceleration and transport of solar energetic particles (SEPs), i.e., ions and electrons accelerated to high energies in solar flares and coronal mass ejections (CMEs). Global modeling of SEP events - including their acceleration in coronal shocks and reconnecting current sheets and their transport to the observer - involves a huge variety of scales ranging from electron scales in the coronal plasma (~1 cm) to the global system size (~1 AU). Thus, multiple plasma simulations are needed to tackle the problem.

In this presentation, we describe the set of simulation tools that are being developed, to understand the acceleration and transport of SEPs in a coherent manner. We discuss the requirements for integration of these tools into coupled simulation models. Our toolbox consists of an MHD simulation describing the global evolution of the bulk plasma during the solar eruption, Monte Carlo simulation tools for the acceleration and transport of electrons and ions, a turbulence transport code describing the evolution of high-frequency MHD fluctuations in the solar wind, and a PiC code describing the fine structure of shocks and current sheets as well as the acceleration of electrons in these environments. We will present case studies where the models are used in concert to obtain a description of SEP acceleration and transport during a solar eruption.

The Atmop project (www.atmop.eu) aims to improve predictions of how space weather affects the trajectories of low-orbiting satellites, via changes to the thermospheric density, which alter the drag on the satellites.
These improved predictions will mainly be provided by an advanced semi-empirical model, as well as creating better proxies for the solar and geomagnetic drivers of the thermosphere-ionosphere system.
However, in order to go beyond the statistical modelling of thermospheric density, and allow better representation of particular or rarely-seen conditions, Atmop also involves work on a physical model of the thermosphere-ionosphere system.

One component of the physical model work is data assimilation - bringing model output closer to reality by regularly incorporating observations, much as realistic weather forecasts are obtained by regularly assimilating observational data into general circulation models of Earth's lower atmosphere.
An important part of data assimilation is controlling the quality of the observational data to be incorporated - ensuring the data is physically reasonable, and that it will not cause problems for the model.

This presentation will present Atmop work on observational quality control in the context of a physical model of the thermosphere-ionosphere system, and discuss the design of the data assimilation scheme.