Научная программа 2-го Российско-Немецкого Совещания
по перспективным исследованиям в вычислительных науках и высокопроизводительных вычислениях
14-16 марта, 2005 г., Штуттгарт, Германия

With the increase of processor performance numerical simulation has become a ubiquitous technology both in science and industry. In both fields it is part of a workflow or a process chain in which it has to be integrated in a seamless way. Sometimes this integration is referred to as e-science. In industry the key words for such integration are information lifecycle management (ILM) and product lifecycle management (PLM). The paper will present the integration of simulation on supercomputers in both scientific and industrial processes.

A boundary value problem for the Navier-Stokes equations for a viscous heat conducting gas in a finite computational domain is considered. The space approximation is constructed with the use of the Bubnov-Galerkin method combined with the method of lines. The parallel realization of this method is discussed for multiprocessor computational system.

The results of the joint project of the Institute of Computational Technologies of Siberian Branch of Russian Academy of Sciences (ICT SB RAS, Novosibirsk, Russia) and the High Performance Computing Center Stuttgart (HLRS, Stuttgart, Germany) are presented. The project is realized within the framework of the activities of the German-Russian Center for Computational Technologies and High Performance Computing (http://www.grc-hpc.de).
The numerical modelling of three-dimensional non-stationary flow in the airbag combustion chamber is done, using the parallelized upwind LU difference scheme. The parallel computations are done in HLRS. The comparison of the results, obtained using different computing platforms, is provided.

The effective iterative algorithms are considered for the numerical realization of the solution of subsidiary differential equations, occurring from using the splitting schemes, for which the convergence theorems are proved and the convergence rates are estimated. The effective algorithm is suggested allowing satisfying identically the grid continuity equation.

Most finite element software is built around a fixed mesh data structure. Therefore, each software package can only be used efficiently for a relatively narrow type of applications. In this talk the speaker will show how a generic mesh interface can be defined such that one algorithm or discretization scheme works on different mesh implementations. The adaptivity and parallelisation included in this generic mesh interface are also presented. Further DUNE modules will be discussed shortly.

Solids respond to a strong mechanical load by plastic deformation, fracture or shock waves. To gain insight into the processes on the atomic scale we have performed molecular dynamic simulations of heavily strained crystals and quasicrystals with up to several million atoms. In these systems the mechanical destruction is mediated by the propagation of elementary modes like dislocations, cracks and shock fronts. Special imaging techniques have been developed to spot the positions of these modes inside a three-dimensional solid and to visualize their dynamics.

The description of DWD-contribution (DWD - Deutscher Wetterdienst) in the work under the EU-Project "An European Flood Forecasting System" (EU-Contract EVG1-CT-1999-00011 EFFS) is represented. The aim of the EU-funded project was the development of a prototype version of a medium-range (up to 10 days ahead) flood forecasting system for the whole of Europe. The brief description of methods of preparing of the input meteorological fields on high performance systems for EU-Project and the complex operational models of an atmosphere used for their construction, and used algorithm of the numerical analysis of high-resolution analyses of 24h precipitation heights on the basis of surface observations for the four flood events in 1994 (Po, November), 1995 (Rhine/Meuse, January), 1997 (Odra, July) and 2002 (Elbe, August) will be given.

For the numerical simulation of large scale CFD and fluid-structure interaction (FSI) problems efficiency and robustness of the algorithms are two key requirements. In this paper we would like to describe a very simple concept to increase significantly the performance of the element calculation of an arbitrary unstructured finite element mesh on vector computers. By grouping computationally similar elements together the length of the innermost loops and the vector length can be controlled. In addition the effect of different programming languages and different array management techniques will be investigated. A numerical CFD simulation will show the improvement in the overall time-to-solution on vector computers as well as on other architectures.
Especially for FSI simulations also the robustness of the algorithm is very important. For the transient interaction of incompressible viscous flows and nonlinear flexible structures commonly used sequential staggered coupling schemes exhibit weak instabilities. As best remedy to this problem subiterations should be invoked to guarantee kinematic and dynamic continuity across the fluid-structure interface. To ensure the efficiency of these iterative substructuring schemes two robust and problem-independent acceleration methods are proposed.

The main line of application of computational methods in General Relativity is concerned with the determination of the waveforms of the gravitational radiation, which is emitted from astrophysical processes. Gravitational wave detectors, currently under construction and calibration, need this information in order to filter out the signals from the noisy background. This contribution describes the basic ideas behind these efforts.

Slender vortices of compressible flow are studied, in particular the deceleration of the axial flow to a free stagnation point on the axis, causing bursting or breakdown of the vortex. Steady, inviscid, compressible, axially symmetric flow conditions are assumed to enable a reduction of the Euler equations for a stream tube of small radius. The angular velocity near the axis is shown to be directly proportional to the axial mass flux, indicating that a deceler-ated axial flow can cause the angular velocity to vanish and breakdown to occur. This behav-ior is reversed in supersonic flow. A breakdown criterion is derived for a Rankine vortex with isentropic and normal-shock deceleration of the axial flow. The results are compared with available experimental data.

The new approach to construction of monotonic nonlinear difference second order schemes, based on the investigation of the differential approximation of a scheme, is presented. One of possible formulas for the definition of approximating viscosity is given, which leads to the coincidence of the constructed scheme with the known Harten's TVD scheme. The known and widely used TVD schemes with other limiters can be also obtained using the presented approach.

Wave propagation over long distances is usually modeled numerically by high order finite difference schemes, compact schemes or spectral methods. The schemes need good wave propagation properties, i.e. low dispersion and low dissipation. In this paper we show that finite volume schemes may be a good alternative with a number of nice properties. The so called ADER schemes of arbitrary accuracy have been first proposed by Toro et. al. for conservation laws as high order extension of the shock-capturing schemes. In this paper we show theoretically and numerically their dispersion and dissipation properties using the method of differential approximation of Shokin. In two dimensions the stability of these ADER schemes is investigated numerically with the von Neumann method. Numerical results and convergence rates of ADER schemes up to 16th order of accuracy in space and time are shown and compared with respect to the computational effort.

Several new possibilities to enhance information capacity of data transmission by integration of several key technologies such as dispersion management, wavelength-division multiplexing and optical regeneration of signals are discussed. Mathematical modelling results may be used for upgrade of existing fiber links and design of new generation of long-haul high-bit-rate communication lines.

This paper describes a case study which deals with the analysis of scalability properties on modern parallel computer architectures in light of a CFD related problem -- the scalable parallel adaption of unstructured grids. It shows how state-of-the-art benchmarking, profiling, and tracing tools can assist authors of parallel CFD applications in making the right design and implementation decisions regarding scalable application performance. A sophisticated platform evaluation framework and a distributed parallel program analyzer are presented.

The geoelectrical model graphs of pole-pole sounding above horizontally stratified medium are calculated. Two methods: method of linear filters and method of direct integration are used. Resistivity profiles above an inclined bed with arbitrary angle of inclination are calculated. The computational procedure of apparent resistivity curves in the area with the thin local embedding is developed.

Vector processors are invaluable tools for high performance numerical simulations due to their high sustained performance. The high efficiency can be attributed to a superior balance between peak performance of the arithmetic pipelines and memory bandwith. In the Teraflop-Workbench project HLRS and NEC investigate how to sustain this performance in the future. We report first results of this cooperation.

Image fusion or registration is central to many challenges in medical imaging today and has a vast range of applications. The purpose of this paper is to give an introduction to intensity based non-linear registration and fusion problems from a variational point of view. To do so, we review some of the most promising non-linear registration strategies currently used in medical imaging and show that all these techniques may be phrased in terms of a variational problem and allow for a unified treatment. A generic registration or fusion method depends on an appropriate chosen distance measure, a regularization, and some additional constraints. The idea of constraints is to incorporate higher level information about the expected deformation. We examine the most common constraints and show again that they may be conveniently phrased in a variational setting. As a consequence, all of discussed modules allow for fast implementations and may be combined in any favorable order. We discuss individual methods for various applications, including the registration of magnetic resonance images of a female breast subject to some volume preserving constraints.

The parametrical analysis of stressed-deformed state of multilayered reinforced nodoid shells on a basis of geometrically linear and nonlinear variants of classical and non-classical theories is made. The influence of structure of reinforcement of a composite material, a cross shift of binding and an order of arrangement of the reinforced layers on behaviour of shells is investigated. The comparison of the numerical solutions obtained by the methods of spline-collocation and discrete orthogonalization is conducted. High efficiency of used numerical methods is shown by example of solution of the boundary value problem for stiff systems of differential equations.

The purpose of a GRID environment construction for scientific GIS problems is considered. The environment is based on parallel computing, centralized warehousing, and decentralized resource control. Some examples of parallel computing are also shown.

The feasibility of particle based simulation methods is shown for powder technological applications like compaction, sintering and filling of a dispersed powder. In contrast to continuum methods this approach automatically takes into account the rearrangements of the grains and predicts structural composition. This allows for a comparison with analytical results in the case of powder compaction and sintering giving new insights into the dynamics of granular materials.

Convection in a spherical shell under the central force field is a classical problem of the fluid dynamics. The applying of a voltage difference between the inner and outer spheres produces the dielectric effect which is responsible for the creation of the central force field. The convective problem has been investigated in the rotating spherical shell for different Prandtl numbers and geometry parameters numerically.

In this Paper we will present a set of modules and plugins developed for the COVISE visualization system which allow setting up interactive CFD simulations. Real objects can be moved in a model and thus provide an easy to use interface to modify the geometry of the simulation.

We address the problem of detecting deviations of binary sequence from randomness, which is very important for random number (RNG) and pseudorandom number generators (PRNG) and their applications to cryptography. Namely, we consider a null hypothesis $H_0$ that a given bit sequence is generated by Bernoulli source with equal probabilities of 0 and 1 and the alternative hypothesis $H_1$ that the sequence is generated by a stationary and ergodic source which differs from the source under $H_0$. We show that data compression methods can be used as a basis for such testing and describe two new tests for randomness, which are based on ideas of universal coding. Known statistical tests and suggested ones are applied for testing PRNGs, which are practically used. Those experiments show that the power of the new tests is greater than of many known algorithms.

We discuss basic optimization and parallelization strategies for current cache-based microprocessors (Intel Itanium2, Intel Netburst and AMD64 variants) in single-CPU and shared memory environments. Using selected kernel benchmarks representing data intensive applications we focus on the effective bandwidths attainable, which is still suboptimal using current compilers. We stress the need for a subtle OpenMP implementation even for simple benchmark programs, to exploit the high aggregate memory bandwidth available nowadays on ccNUMA systems. If the quality of main memory access is the measure, classical vector systems such as the NEC SX6+ are still a class of their own and are able to sustain the performance level of in-cache operations of modern microprocessors even with arbitrarily large data sets.

"Virtual Organizations" belong to the key concepts in the Grid computing community. They are currently evolving from basically static to dynamic solutions that are created ad-hoc in reaction to a market demand. This paper provides a definition of "dynamic Virtual Organizations" in order to assess specific challenges of an abstract collaborative engineering scenario. The paper concludes with a description of an evolving architecture enabling such dynamic virtual organizations.

The performance of algorithms dependents on a whole bunch of parameters, not only frequency of the processor but also its architecture, bandwidth and different latencies for getting data. Also the implementation of the algorithm is essential. We try to identify some important parameters by the analysis of the delivered performance of some typical algorithms and to show the differences between architectures.

This paper presents the design and implementation of the Science Grid Modeling Laboratory (SGM-Lab), an automated parametric modeling system for performing complex dynamically-controlled parameter studies. Nowadays, simulation programs are used not only in research but also during the development of products, often to optimize their quality. Typically, this involves repeated execution of the simulation codes, whereby for each run some of the input data is varied. As a result, many different jobs have to be launched and a huge amount of output data has to be administered. A grid environment can provide, and enable the exploitation of the necessary resources for this computation. However, in order to be able to use a grid environment effectively, tool support is required to automatically generate the parameter sets, issue jobs, control the successful operation and termination of jobs, and collect results. Support is also needed to generate new parameter sets based on previous results in order to obtain a functional optimum, after which the parameter study should terminate. The SGM-Lab software described in this paper offers a unified framework for such large-scale optimization problems.

Canonical homogeneous turbulent flows like simple shear and irrotational stagnation-point turbulence have been and are being simulated numerically at sufficiently high Reynolds numbers to allow turbulence modelling that bridges the range of Reynolds--number dependence. But direct simulations of canonical channel flows are still struggling to penetrate the high--Reynolds regime and to produce the logarithmic law found in that regime. Moreover, the basic experimental flow used to study homogeneous turbulence, namely grid--generated turbulence, has remained untreated by direct simulation. We present preliminary results toward the goal of simulating high--Reynolds--number turbulence in these flows, as well as of the more complicated flows obtained by combining the effects isolated in the canoncal flows. In particular, channel turbulence can be produced and enhanced using grids, wall roughness, contractions, and other fixed devices that promote turbulence. We present lattice Boltzmann DNS for plane channel, grid-in-channel, and linearly contracting channel geometries. We show the high relevance of these flows to ongoing research in turbulence modeling and estimate the required resources for corresponding high--Reynolds--number simulations. The potential of lattice Boltzmann methods to provide an optimal solver on vector--parallel and highly parallel architectures is shown.