Scientific program of the 3rd Russian-German Advanced Research Workshop on Computational Science and High Performance Computing
July 23-27, 2007, Novosibirsk, Russia

organize

The 3rd Russian-German Advanced Research Workshop
on Computational Science and High Performance Computing

sponsored by
Deutsche Forschungsgemeinschaft (German Research Foundation)

July 23-27, 2007

Institute of Computational Technologies SB RAS

Akademgorodok, Novosibirsk, Russia

Program committee

Co-chairmen:

Academician Yu. I. Shokin (ICT SB RAS)
Prof. M.M. Resch (HLRS)

Members of program committee:

Prof. M.P. Fedoruk (ICT SB RAS)
Corresponding member of RAS V.V. Shaidurov (Institute of Computational Modelling SB RAS, Krasnoyarsk, Russia)
Prof. I.V. Bychkov (Institute of System Dynamics and Control Theory SB RAS, Irkutsk, Russia)
Prof. B.Ya. Ryabko (SibSUTI, ICT SB RAS)

Organizing committee

Chairman:

Prof. M.P. Fedoruk (ICT SB RAS)

Members of organizing committee:

V.B. Barakhnin (ICT SB RAS)
D.L. Chubarov (ICT SB RAS)
U. Küster (HLRS)
N.N. Kuznetsova (ICT SB RAS)
Yu.I. Molorodov (ICT SB RAS)
V.A. Pestunov (ICT SB RAS)
N.Yu. Shokina (HLRS)
V.N. Yarutkin (ICT SB RAS)
V.A. Yasakov (ICT SB RAS)
A.V. Yurchenko (ICT SB RAS)

Working language: English

Program of the Workshop

The Institute of Computational Technologies SB RAS provides expertise in developing mathematical and numerical models in a wide range of disciplines including mechanical engineering, physics and ecology. To cater for the diverse needs of the users within the institute a centralized computational facility is being developed. Several avenues of development are pursued at once. The Institute is introducing new hardware to increase the raw compute power, runs an educational program and works to improve the utility of the existing computational resources for fundamental research and industrial collaborations. We present an overview of the computational resources currently available and describe several case studies that describe the role the computing facility plays in some of the research projects the Institute is currently working on.

Grid computing has become a standard technology to connect computers and instruments as well as data and software. This allows creating virtual pools of resources that can easily be used. As part of this so called service oriented architectures (SOA) have evolved that claim to provide the user with a service while hiding the complexity of hardware and software. HPC has to adapt to this new scenario. Supercomputing is in itself a service and the user is mainly interested in this service and not in the details of the hardware. The paper will describe how HPC can be defined as a service in a Grid environment.

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 parallel realization of the problem on mathematical modelling of three-dimensional electromagnetic fields in inhomogeneous domains in frequency domain is presented. The computational schemes are investigated in domains, which are inhomogeneous with respect to physical (electroconductivity) and geometrical (different scales of individual fragments of a modelling domain) characteristics, in the presence of field sources changing geometrical position with respect to interfragmentary boundaries, which separate individual subdomains. The three-dimensional vector Helmholtz equation is approximated on inhomogeneous tetrahedral grid with local grid refinements using the vector finite element method on the vector basis functions of high order. The basis functions (edge-elements) for calculating the electric field guarantee continuity of tangential components of the field on interelement and interfragmentary boundaries. For exact calculation of a jump of electric field normal component the special procedure for taking into account a weak divergence on basis functions of high order is realized. The problem on high-frequency induction logging isoparametric soundings (the VIKIZ-method in Russia) in a domain with complicated geometry is solved. The parallelization is done by formulating of n independent problems, where n is a number of probe positions, and associating each of these problems with a separate processor of computing system. The calculation results are presented, obtained on NEC SX-8 cluster.

The inverse problem of the mathematical theory of tides is considered in the form of defning a boundary values on liquid parts of boundary. The direct and conjugate equations of shallow water are closed by the observation data on the function of sea level (free surface elevation) on a part of the boundary. The iterative algorithm is used for solving this complete problem in connection to tides of World Ocean.

After decades of increasing per processor performance with small architectural changes progress is now facing a handicap in the stagnation of the processors frequencies. The chip densities and sizes are still enlarging and used for increase of the number of cores. What bottlenecks we have to be aware? What are the implications for numerical algorithms? Are there foreseeable rules for implementations?

Catastrophic tsunamis that flooded the ocean coast in the past had taken many human lives and destroyed the infrastructure of the coastal areas in the Pacific and elsewhere. Recent tragic events in the Indian Ocean motivated governments to develop new and improve the existing tsunami warning systems capable to mitigate the impact of catastrophic events. In the future, regional tsunami warning systems will be integrated into a network including both the systems currently being developed and the existing warning systems (in the past, tsunami warning systems had been deployed to protect the coastal areas in Japan, USA, Russia, Australia, Chile and New Zealand). The present study, conducted by the Institute of Computational Technologies in collaboration with other research institutes in Novosibirsk, is directed to the design of a new generation of the tsunami warning system for the Pacific coast of Kamchatka. The main purpose of the present study is to develop the computational methodology allow us to build a database of potential tsunamigenic sources that impose the imminent tsunami risk for the eastern coast of Kamchatka Peninsula. In the first stage of the project, a set of basic model sources of tsunamigenic earthquakes is defined. These model sources are used to calculate the initial water elevation in the source area that are used as initial conditions for dynamic modeling of tsunami propagation in the selected geographical region near the Kamchatka east coast. The next stage is the modeling of propagation and transformation of tsunami waves on the way from the source area towards the coast. This information is presented as a decision support system intended for use by persons on duty who are responsible for initiating tsunami mitigation procedures such as evacuation of people and sending ships away from the dangerous harbors and bays. The project involves numerical solution of a large number of instances of wave hydrodynamics problems. At the same time, interpretation of the results requires non-trivial postprocessing and development of a specialized information systems. Numerical calculation of tsunami propagation in the ocean with a real bathymetry for multiple combinations of the model earthquake sources with high spatial resolution constitutes the major part of the computational requirements of the project. The amount of computation resources needed for numerical modeling of tsunami propagation implies the use of high performance computers and may require adaptation of numerical algorithms to specific computing platforms. In the future, this approach may be extended to other tsunamigenic areas in the Pacific and the Indian oceans. In the latter case, we plan to consider the problem of defining the extension of the coastal zone that is the subject for flooding by tsunami waves. This would require to perform run-up calculation. To achieve the necessary accuracy in reproducing of tsunami behavior in the coastal areas, we will need to take into account such features of the coastal area as small rivers, lakes and swamps. This will significantly increase the amount of computational resources needed both during the development of numerical models and algorithms and in the production runs.

In this paper we present numerical simulation.s for PEM (Polymer Electrolyte Membrane) Fuel Cells. Hereby, we focus on the simulation done in 3d using modern techniques like higher order discretizations using Discontinuous Galerkin methods, local grid adaptivity, and parallelization including dynamic load-balancing. As a test case for the developed software we simulate the two-phase flow and the transport of species in the cathodic gas diffusion layer of the Fuel Cell. Therefore, from the detailed model presented in [1] we derive a simplified Model Problem presented in Section 1. In Section 2 one finds a few notes on the discretization schemes that were used for the simulation including comments on adaptation and parallelization. In Section 3 the results of an adaptive, parallel simulation in 3d are presented.

1. Steinkamp K, Schumacher J O, Goldsmith F, Ohlberger M, Ziegler C (2007) A non-isothermal PEM fuel cell model including two water transport mechanisms in the membrane. In: Preprint series of the Mathematisches Institut of the Universit\auml;t Freiburg

A series of numerical models of free turbulent fows developed by authors are considered in this paper. These numerical models are based on the modern semi-empirical models of turbulence.

The study of fluid flow inside compliant vessels, which are deformed under an action of the fluid, is important due to many biochemical and biomedical applications, e.g. the flows in blood vessels. The mathematical problem consists of the 3D Navier-Stokes equations for incompressible fluids coupled with the differential equations, which describe the displacements of the vessel wall (or elastic structure). We study the fluid flow in a tube with different types of boundaries: inflow boundary, outflow boundary and elastic wall and prescribe different boundary conditions of Dirichlet- and Neumann types on these boundaries. The velocity of the fluid on the elastic wall is given by the deformation velocity of the wall. In this publication we present the mathematical modelling for the elastic structures based on the shell theory, the simplifications for cylinder-type shells, the simplifications for arbitrary shells under special assumptions, the mathematical model of the coupled problem and some numerical results for the pressure-drop problem with cylindrical elastic structure.

The paper addresses numerical modeling of electric and magnetic fields in fiber media generally described by Maxwell equations. Complex microstructures and composite materials in modern optics devices require special techniques to be modeled. These peculiarities are all considered in FVTD method proposed. High computational complexity inherent to multidimensional computations is resolved through parallel computations. For this purpose the decomposition of nonstructured triangular grids and MPI programming model are implemented. The resulting efficiency graphics obtained on HLRS cluster (NEC Xeon EM64T Cluster) show good scalability. In particular case, the simple fiber waveguide can be described by the simplified mathematical model. Nevertheless the computational complexity increases when femtosecond laser inscription with close to critical input power are considered. In order to parallelize these computations the parallel Thomas algorithm for solving tridiagonal linear systems was implemented. The obtained efficiency graphics are discussed.

High Reynolds number flows are still challenging problems for large-eddy simulations (LES) due to thin small-scale structures e.g. in the near wall regions and often transitional boundary layers which have to be resolved. For this reason, the prediction of high Reynolds number airfoil flow over the entire geometry using LES still requires huge computer resources. To remedy this problem a hybrid zonal RANS-LES method for the flow over an airfoil in high-lift configuration at Re_c=10⁶ is presented. In a first step a 2D RANS solution is sought, from which boundary conditions are formulated for an embedded LES domain, which comprises the flap and a sub-part of the main airfoil. The turbulent fluctuations in the boundary layers at the inflow region of the LES domain are generated by controlled forcing terms, which use the turbulent shear stress profiles obtained from the RANS solution. In the second part of the paper a large-eddy simulation of the flow around an airfoil consisting of a slat and a main wing is performed at a Reynolds number of 1.4×10⁶ based on the freestream velocity and the clean chord length to identify the flow phenomena generating slat noise. The freestream Mach number is Ma=0.16 and the angle of attack is 13⁰. A computational mesh with about 55 million cells is used to resolve the turbulent scales in the boundary layers and within the slat cove region. Results are presented for both the turbulent flow field obtained from the LES and the acoustic field obtained with a computational aeroacoustics method based on acoustic perturbation equations.

Efficient statistical tests, e.g. recently suggested "Book Stack" test, are successfully applied to detect deviations from randomness in bit sequences generated by stream ciphers such as RC4 and ZK-Crypt, as well as by block cipher RC6 (with reduced number of rounds). In case of RC6 a key recovery attack is also mounted. The essence of the tests is briefly described. The experiments data are provided.

Delivering high performance on contemporary high performance computing architectures becomes more and more challenging owing to changing trends in computer hardware but also due to the incorporation of more complex physical models. Results from low level benchmarks and flow simulations using lattice Boltzmann approaches are reported for advanced HPC systems (NEC SX-8 vector parallel computer), commodity HPC clusters (based on Intel Woodcrest CPUs and Infiniband interconnect) and special purpose hardware (IBM Cell processor).

The paper presents a new class of adaptive and sequentially guaranteeing PPS-methods based on partitioning parameter sets, for computing optimal (exact) component-wise bounds of the solution sets to interval linear systems with square regular matrices.

Free surface problems in fluid mechanics are of a great applied importance, and that's why they rouse many researchers to excitement. Numerical simulations of such problems, using the so-called meshless methods, become more and more popular today. One of the subsets of these methods is the class of meshless particle methods, which don't use any mesh during the whole numerical simulation process, and therefore allow solving the problems with large deformations and with failure of problem domain connectivity. These reasons cause great popularity of these methods in the sphere of numerical simulation of free surface problems. One of the meshless particle methods is the Smoothed Particle Hydrodynamics (Gingold RA, Monaghan JJ (1977) Mon Not Royal Astron Soc 181:375-389). Numerous computations, mainly of free surface problems, carried out by different scientists, using the mentioned method, proved its doubtless efficiency in receiving the high-quality kinematic images of ideal as well as viscous fluid flows. However, it has a considerable drawback - it doesn't allow receiving satisfactory images of pressure distribution. Recently, with the purpose to deliver such drawbacks, the ISPH (Incompressible Smoothed Particle Hydrodynamics) method has been developed (Cummins SJ, Rudman M (1999) J Comp Phys 152:584-607; Shao S, Lo EYM (2003) Adv Water Res 26:787-800), which is used for simulation the incompressible fluid flows, as its name shows. It uses a split step scheme for integration of basic equations of fluid dynamics. Comparative analysis of computational results, obtained by the methods, mentioned above, shows, that the ISPH gives low-grade kinematic images of fluid flows in comparison with the classical SPH, however, it allows obtaining much better images of pressure distribution, and it will give an opportunity to compute the hydrodynamics loads in future.

The design and execution of complex scientific applications in the Grid is a difficult and work-intensive process which can be simplified and optimized by the use of an appropriate tool for the creation and management of the experiment. We propose SEGL (Science Experimental Grid Laboratory) as a problem solving environment for the optimized design and execution of complex scientific Grid applications. SEGL utilizes GriCoL (Grid Concurrent Language), a simple and efficient language for the description of complex Grid experiments.

The paper presents the fundamentals of a service-oriented architecture (SOA), its layers, and the associated types of architectural decisions. The method for service-oriented modeling and the characteristics of activities from a service consumer and provider perspectives are described. This method guarantees necessary requirements for analysis and design in order to determine three fundamental aspects of a service-oriented architecture: services, flows, and components that realize the services. SOA is an architectural style, which has a goal to achieve a loose coupling among interacting software agents. A service is a unit of work done by a service provider to achieve desired final results for a service consumer. Both provider and consumer are roles played by software agents on behalf of their owners. A template is described, which can be used for architectural decisions in each of the SOA layers. For service identification it is important to combine three approaches of top-down, bottom-up, and cross-sectional analysis of problem modelling. The main activities of specification and realization of services are considered.

Telemedicine is the using of achievements of telecommunication technologies in the public health service. Telemedicine allows improving the efficacy of treatment and diagnosis on a new level. With the aid of such technologies you can render a great qualified help to a remote patient in the best medical institutions of our country as well as of the whole world. Doctors can make a diagnosis using as the base received through the e-mail or the Internet X-ray pictures, computer tomograms, electrocardiograms, brainwaves or other findings of laboratory or instrumental patient's examinations. In connection with the fact that the essential part of specialists in different fields of medicine work in specialized medical centers of big cities, so it provoked some centralization of attendance. However, the achievements of telemedicine remove the necessity of direct physical presence of a specialist. Modern telemedical ways allow realizing removed advisories of doctors and their patients locating in far regions. By the way, for realizing of consultation of any ill person a physician can repose not only his own experience. Thank to telecommunication technologies physicians and specialists can listen to lectures of famous scientists of the most actual problems of the public health service and medical science, to keep professional contacts with global scientific centers, and also with their colleagues from neighboring district hospitals or from principal specialists of regional center. The possibility of using of technologies of videoconferences is very attractive which permits the real communication in the regimen of video transmission.

Dynamic interaction between a stratosphere and troposphere, in particular, connected with planetary waves, can render essential influence on variations of tropospheric circulation with time scales from several days about several months. Climatology of extratropical stratospheres it is defined by intensity of a polar vortex, which is connected, with gradients rate of temperature. Recently there is decrease of temperature of the top layers of an atmosphere, contrast between a tropical and polar stratosphere and, accordingly, a temperature gradient decreases. Therefore the question on how changes of a polar vortex influence a condition of the bottom layers of an atmosphere represents essential interest. So, for example, in several works, using daily and monthly average data has been established communication between variations surface pressure, strength of polar vortex and tropopause altitude. Increasing of polar vortex connected with low tropopause over Iceland and high tropopause over Arctic. High tropopause altitude is connected with strength and twisting of column of air and small drop of surface pressure in polar region. However, we have no deep understanding of mechanism of interaction of atmosphere and stratosphere. This report deals with modeling research of the problem using atmosphere general circulation model.
The work was supported by the RFBR grant No. 05-05 - 64989

Since a couple of years, multi-core chips get more and more common in high performance computing (HPC), and we expect to see systems with more than 100000 cores even in reasonably sized HPC systems. This will increase the peak performance of such computers to Petaflops and more. The only remaining question is: Are we prepared to efficiently use such machines? Do we already know how to program these computers, how to detect bottlenecks and how to improve efficiency? The talk will address these topics and describe actual trends and perspectives.