Report run on: 11/09/2000
12/10/2000
Session: Graphics and Visualization
Chair: P. Kacsuk
14:30
CA1 CGL: A Computational Geometry Library An overview of an object-oriented class library, CGL, for computational geometry is presented. The library has recently emerged in response to the continued growth and application of the discipline Computational Geometry. The structure and class hierarchies of CGL are presented and how CGL forms part of a larger computational modelling class framework, CML, and how the library interfaces with other modellers and is used by application-based libraries.
G.M. Seed, Dept. of Mechanical and Chemical Engineering, Edinburgh, UK
14:45
CA2 Visualization of Seismic Data in Geophysics and Astrophysics Seismic data obtained from surface observations of the Earth and the Sun has enormous importance in increasing our knowledge of the internal structure of these objects. By making special assumptions (e.g. spherical symmetry) it is possible to use analytic tools or relatively simple numerical calculations to analyze the tomographic data. In both cases, however, the real problem is inherently three dimensional and to resolve detailed structure enormous amounts of data are required. We show in this paper how visualization can aid the process of interpretation. The spherical geometry of such objects, and the huge variations in physical quantities with depth, present formidable challenges to current visualization systems. We present a major case study using AVS/Express to visualize some of the largest geophysical simulations ever created and show how we intend to extend these techniques to astrophysics.
J. Leng(1), J. Brooke(1), T. Hewitt(1) and H. Davies(2)
(1) CSAR, University of Manchester, UK
(2) Dept. Earth Sciences, University of Liverpool, UK
15:00
Computers graphics are very extensive and it is acquiring more and more relevance in generating more or less realistic scenes, but it also allows to deep inside them like one more element of the scene or to integrate virtual scenes with real one, or to obtain stereoscopic vision. Virtual Reality offers all of these possibilities.
So we have developed one application that employ Radiosity to generating photorealistic images. It also allows stereoscopic vision and movement of the viewer in order to achieve more realism for the virtual word.
CA3 Scenes Visualization in Virtual Reality Systems Using the Radiosity Method
J.I. Benavides, M.A. Ortiz, F.J. Delgado, M.J. Cuevas, Escuela Politécnica Superior, Spain
15:15
CA4 3D Trees Modelling In this work we present an application to facilitate the selection and inclusion of previously generated trees, in three-dimensional scenes.
The application consists of a database where the trees are stored. The trees are generated in different formats. And the application includes a user interface that allows the selection of the desired tree, its visualisation and a description of its habitat.
J.I. Benavides, M.A.Ortiz, J.M. Canto, F. Serrano, Escuela Politécnica Superior, Spain
15:30
CA5 Parallelized Image Processing Algorithms for HPC Systems In this paper are shown some aspects regarding the importance of HPC solutions in Image Processing. According this goal, are shown some n-dimensional extensions of the definition of some Image Processing operators, being analyzed the sequential solution in direct comparison with the parallelized ones.
P. Mitrea(1), D. Grigoras(2)
(1) Software ITC SA Cluj-Napoca, Romania
(2) Techincal University "Gh. Asachi" Iasi, Romania
Session: Tools for High Performance Computing I
Chair:
R. Wismüller
14:30
CB1 The UNICORE Grid Infrastructure UNICORE (Uniform Interface to Computer Resources) is a software infrastructure to support seamless and secure access to distributed resources. UNICORE allows uniform access to different hardware andsoftware platforms as well as different organizational environments. Based on the abstract job model it offers services for security, translation of abstract jobs into real batch jobs for different target systems, and a public key infrastructure. This paper describes the UNICORE architecture and provided services.
M. Romberg, Research Center Jülich, Germany
14:45
CB2 Application Servers in Polish Computing Grid The article describes an idea of dedicating single supercomputers to specific applications. This idea has been introduced into the Polish National Computing cluster. Allocating applications to specific supercomputer architecture can increase utilisation of the systems and decrease computer administration costs. A proposition of configuration in the Polish National Cluster is also discussed. The application servers concept is under development and is going to be finished next year.
M. Kupczyk, N. Meyer, M. Stroiński, P. Wolniewicz, Poznań Supercomputing and Networking Center, Poland
15:00
CB3 Integrated Access Control and Dynamic Event Reporting Facilities in the VALIS Intrusion Detection System This paper presents an approach to the security of operating systems, which is based on the integration of mechanisms for active protection and unauthorized access detection. In its main part a brief overview of the VALIS IDS system developed in the PSNC is shown. Then a more detailed description of the design and architecture of one of its modules - the SCM module is given, which realizes the active protection and dynamic event reporting facilities at the system call level. Final conclusions are given at the end of the paper.
M. Chmielewski, A. Gowdiak, S. Fonrobert, N. Meyer, T. Ostwald, M. Stroiński, Poznań Supercomputing and Networking Center, Poland
15:15
CB4 Portable Library of Migratable Sockets Efficient load balancing is essential for the development of parallel distributed computing. Many parallel computing environments use TCP or UDP through the socket interface as a communication mechanism. This paper presents design and development of a prototype implementation of a network interface which may keep communication between processes during process migration. This new communication library is a substitution of the well-known socket interface. It is implemented in the user--space; it is portable, and no modification of user applications are required. The TCP/IP is applied for internal communication what guarantees relatively high performance and portability.
Keywords: distributed computing, load balancing, process migration, Dynamite, sockets.
M. Bubak(1,2), D. Żbik(1), D. van Albada(3), K. Iskra(3), P. Sloot(3)
(1) Institute of Computer Science, AGH, Poland
(2) ACC CYFRONET AGH, Poland
(3) Informatics Institute, Universiteit van Amsterdam, The Netherlands
15:30
CB5 Irregular and Out-of-Core Computing on SGI Origin 2000 This paper presents the lip runtime system which was elaborated to support parallelization of irregular and out-of-core computations. Functions for handling irregular data were developed using the similar approach as in the CHAOS library while out-of-core parallelization is based on the idea of in-core section. Functions for out-of-core data are implemented with capabilities provided by MPI-IO. Preliminar results of performance tests for a generic irregular out-of-core program on SGI Origin 2000 are presented.
Keywords: irregular problems, out-of-core computing, I/O operations, runtime support.
M. Bubak(1,2), P. Łuszczek(3), M. Malawski(1), K. Zając(1)
(1) Institute of Computer Science, AGH, Poland
(2) ACC CYFRONET AGH, Poland
(3) Dept. of Computer Science, Univ. of Tennessee, USA
Session: Molecular Dynamics Simulation
Chair: B. Szymański
16:00
CC1 Application of Pfortran and Co-Array Fortran in the Parallelization of the GROMOS96 Molecular Dynamics Module After at least a decade of parallel tool development, parallelization of scientific applications remains a significant undertaking. Typically parallelization is a specialized activity supported only partially by the programming tool set, with the programmer involved with parallel issues in addition to sequential ones. The details of concern range from algorithm design down to low-level data movement details. The aim of parallel programming tools is to automate the latter without sacrificing performance and portability, allowing the programmer to focus on algorithm specification and development. We present our use of two silimar parallelization tools, Pfortran and Cray's Co-Array Fortran, in the parallelization of the GROMOS96 molecular dynamics module. Our parallelization started from the GROMOS96 distribution's shared-memory implementation of the replicated algorithm, but we used little of that existing parallel structure. Consequently, our parallelization was close to starting, with the sequential version. We found the intuitive extensions to Pfortran and Co-Array Fotran helpful in the rapid parallelization of the project. We present performance figures for both the Pfortran and Co-Array Fortran parallelizations.
P. Bała(1,2), T. Clark(3), L.R. Scott(3)
(1) Interdisciplinary Centre for Mathematical and Computational Modelling, Warsaw Univ., Poland
(2) Faculty of Mathematics and Computer Science, N. Copernicus University, Toruń, Poland
(3) Dept. of Computer Science, Univ. of Chicago and Computation Institute, USA
16:15
The Gaussian98 program run on SGI Origin was used for ab initio calculations of some molecules, such as biphenyl derivatives, cyanoanilines and polyacetylenes. Geometry optimization in ground and excited state as well as energy barrier estimation and vibrational frequencies calculations were performed. The results were compared with information about excited states of the molecules obtained from laser induced fluorescence spectra of molecules cooled in molecular beam. The results were also compared with IR or Raman spectra or UV absorption spectra.
CC2 Ab initio Calculations and Interpretation of the Fluorescence Excitation Spectra of Jet-cooled Molecules
K. Pirowska, P. Kolek, J. Najbar, Faculty of Chemistry, Jagiellonian University, Kraków, Poland
16:30
The Structure of Bismuth-Silicate and Bismuth-Germanate Glasses: a Molecular Dynamics Study We report on the results of molecular dynamics (MD) simulations of the structure of bismuth-silicate and bismuth-germanate glasses with bismuth oxide concentration in the range from 15% mol to 60% mol. The simulations have been performed in the isobaric- isothermal ensemble, using two-body interaction potentials. The set of the potential parameters was constructed as a suitable combination of the parameters, which were previously proposed for pure silicone oxide [Soules T F 1982 J. Non-Cryst. Solids 49 29], germanium oxide [Nanba T et al. 1994 J. Non-Cryst. Solids 177 131], and bismuth [Abrahamson A A 1969 Phys. Rev. 178 76]. All the systems were initially prepared as well equilibrated hot melts, and then slowly cooled down to 300K. The structural information from the MD simulations was obtained from radial and angular distribution functions for the short-range order, and for the medium-range order from analysis of anion-cation rings. The ring analysis has been performed using a new redundancy aware algorithm [Mancini G 1997 TASK Quart. 1 89].
CC3
A. Witkowska(1,2), J. Rybicki(1,2), M. Białoskórski(1), G. Mancini(3), S. Feliziani(3)
(1) Dept. of Solid State Physics, Technical University of Gdańsk, Poland
(2) TASK Computer Centre, Gdańsk, Poland
(3) INFM UdR Camerino, Universita di Camerino, Italy
16:45
CC4 Computer Simulations in Molecular Modelling The molecular modelling is oriented on the computer aided drug design discipline. The large-scale calculations are necessary to analyse the static and dynamic forms of biologically active systems.
A model of immunoglobulin light chain dimer was used as the object of analysis. This protein is partially responsible for the activity of human immunological system. It represents a very interesting model of function-related structural changes in proteins, which can be applied to computer aided drug design.
M. Król, M. Nowak, I. Roterman, Dept. of Biostatistics and Medical Informatics, Collegium Medicum, Jagiellonian University, Cracow, Poland
17:00
Mixing and Droplets Coalescence in Immiscible Fluid. 3-D Dissipative Particle Dynamics Model We employ dissipative particle dynamics (DPD) in three dimensions to simulate the Rayleigh-Taylor (R-T) mixing for two superimposed, viscous, immiscible and compressible fluids. The two antagonistic processes: droplets break-up and droplets coalescence, influence the speed of mixing. We show that DPD simulation results are in good agreement with theoretical predictions. Then we concentrate on the coalescence problem in equilibrium. The temporal growth of the average droplet size R(t) is govern by the power law R(t)~t^b. The value of the scaling factor b varies with time and depends on the type of physical mechanism, which is dominant in a moment of time. Initially, the value of b is equal to 1/3 for small R(t), then increases to 1. Eventually, for R(t)>RH, where RH is the hydrodynamic length, the value of b is equal to 2/3. We detected also an intermediate b~1/2 regime between b~1/3 and b~1 regimes, observed before only for 2-D systems. The b~1/2 regime becomes dominant for sufficiently small kinematic viscosity. The simulations were performed on the SGI/Origin 2000 system by using 3*10^5 DPD particles.
CC5
K. Boryczko(1), W. Dzwinel(1), D.A.Yuen(2)
(1) Institute of Computer Science, AGH, Poland
(2) Minnesota Supercomputer Institute, University of Minnesota, USA
Session: Tools for High Performance Computing II
Chair:
M. Romberg
16:00
CD1 On the Bisection Bandwidth of Origin2000 The bisection bandwidth is an important performance parameter of large parallel message passing systems in particular for transposition of data between different computational phases of a program but also for boundary exchange of grid oriented problems. The bisection benchmark of mpptest for MPI_Send/MPI_Recv constructs has been used to investigate this parameter for the MPI programming platform on large SGI Origin2000 systems. No severe bottleneck has been found in the system structure. For small messages the start-up time is growing with the system size during a bisection communication. For large messages the runtime of bisection communication is about 20% slover than for a single message exchange. There are cases where an additional factor of 2 can be lost. In addition, there is a special mode of data transmission that runs bisection benchmark much faster than the standard mode with default MPI parameters.
Key Words: message passing, performance analysis, benchmarking, MPI, ccNUMA architectures, bisection bandwidth.
H. Mierendorff(1), H. Schwamborn(1), R. Vogelsang(2)
(1) Institute for Algorithms and Scientific Computing (SCAI), GMD, Germany
(2) Silicon Graphics GmbH, Grasbrunn, Germany
16:15
CD2 PC Clusters - Parallel Supercomputers of the Future? We have investigated the current state of the art of PC clustering, and performed benchmarks with a view to supplementing (or even replacing) our 86 processor Cray/SGI Origin 2000. The implications of such a strategy on the user community and the systems staff are presented and discussed.
D.J. Baker, Oxford Supercomputing Centre, Oxford, UK
16:30
CD3 Towards Extension of an OCM-based Tool Environment There is a high demand for tools supporting parallel programming. These comprise several classes of tools, e.g. debuggers, performance analyzers, visualizers, etc., each class providing a well defined set of services. A well designed tool environment should support monitoring of parallel applications developed using different programming models and offer a wide range of tools capable of cooperating with each other. This paper presents the recent development of the OCM-based environment of on-line tools and outlines perspectives for further research.
Keywords: parallel programming, monitoring, on-line tools, interoperability, MPI, OMIS.
M. Bubak(1,2), W. Funika(1). B. Baliś(1), R.Wismüller(3)
(1) Institute of Computer Science, AGH, Kraków, Poland
(2) ACC CYFRONET AGH, Kraków, Poland
(3) LRR-TUM, Technische Universitaet München, Germany
16:45
CD4 Predicting Job Execution Times in the Grid Information discovery is vital to grid computing. Dynamic, hard to predict nature of grid makes it very hard to manage and schedule. One of difficulties which grid schedulers have to approach is lack of efficient information about user jobs. Knowing the job execution times is not a trivial task. Usually, it is quite complicated to get the processing time requirements from the user. Moreover, the user usually gives the upper bound of this time, if at all. Obviously, having those times would make it much easier and better to schedule jobs in the grid environment.
The paper describes a methodology for finding efficient descriptors of job classes (so called templates) and method of using those templates for a dynamic job execution time prediction in a grid environment. The method is mostly based on the historical data collected by the resource management system. The way in which the method will be used to improve local scheduling systems is also presented. Although the method will be used in a global grid, the preliminary results shown in the paper are based on the SGI machines installed in Poznań and Gdańsk.
K. Kurowski, J. Nabrzyski, J. Pukacki, Poznań Supercomputing and Networking Center, Poland
17:00
CD5 Creating Java to Native Code Interfaces with Janet Extension As Java is being considered an appropriate environment for high performance computing, the interest arises in combining it with existing code written in other languages. Portable Java to native code interfaces may be developed using the Java Native Interface (JNI). However, as a low-level API it is rather inconvenient to be used directly thus the higher level tools and techniques are desired. We present JANET - the highly expressive Java language extension enabling convenient creation of powerful and efficient Java to native code interfaces.
M. Bubak(1,2), D. Kurzyniec(1), P. Łuszczek(3)
(1) Institute of Computer Science, AGH, Kraków, Poland
(2) ACC CYFRONET AGH, Kraków, Poland
(3) Dept. of Computer Science, University of Tennessee, USA
13/10/2000
Session: Finite Element Modeling
Chair: C.J.K. Tan
14:00
CE1 Numerical Simulations of Fluid Mixing and Optimization of Mixers
 A computational tool for fast simulation of laminar fluid mixing, capable of dealing with industrial mixer geometries is presented. This method uses extensive computations to create a sparse mapping matrix (set of matrices) that describes fluid transport between the cells covering the flow domain. These extensive computations necessitate parallel processing, which is easily realizable due to the huge number of independed steps involved. Then, mixture evolution on a (large) time step is computed using matrix-vector multiplications. We evaluate the capability of the technique to cope with industrial mixer geometries. This forms the basis for optimization of mixer design.
O.S. Galaktionov, P.G.M. Kruijt, P.D. Anderson, G.W.M. Peters, H.E.H. Meijer, Dutch Polymer Institute, Materials Technology, Eindhoven University of Technology, The Netherlands
14:15
CE2 3D Computer Simulations of Thermosetting Materials Molding FEM and FVM based computer simulation programs are very useful tools used in the investigation of various reactive molding processes.The selected results of 3D computer simulation of reactive molding process for manufacturing of voltage transformers is presented. The simulation has been based on FLUENT computer code. Due to complexity of the process, curing kinetics and viscosity models were implemented in calculations. Silicon Graphics server ORIGIN 2000 was used to obtain fast and accurate calculations. Results obtained allow to optimise process parameters and produce components with better durability and quality.
R. Sekuła, P. Saj, T. Nowak, K. Kaczmarek, ABB Corporate Research, Kraków, Poland
14:30
CE3 Parallel Finite Element Modeling on PC-Based Clusters with SMP Nodes The paper presents a concept and implementation of the NuscaS object-oriented software environment for the parallel FEM modeling on clusters of workstations. One of key directions of extending the functionality of NuscaS is efficient utilization of SMPs as cluster nodes. In the paper, we discuss two schemes of implementing FEM computations on PC-based clusters with SMP nodes. They are oriented on using either the pure message- passing model of parallel programming or a mixture of the message-passing and shared-memory models. Preliminary performance results for a node with two Pentium Celeron processors are given.
R. Wyrzykowski, T. Olas, N. Sczygiol, Technical University of Częstochowa, Poland
14:45
CE4 Explicit and Implicit Parallel Programming Models for Finite Element Method on Multiprocessor Systems We compare three parallel programming models applied to an adaptive finite element code for solving nonlinear simulation problems on unstructured grids with the preconditioned GMRES method. They use the domain decomposition, the functional decomposition and both respectively.
J. Płażek(1), J. Kitowski(2,3), K. Banaś(1)
(1) Section of Applied Mathematics FAPCM, Cracow University of Technology, Poland
(2) Institute of Computer Science, AGH, Kraków, Poland
(3) ACC CYFRONET AGH, Kraków, Poland
15:00
CE5 High Quality 2D Mesh Generation Based on Delaunay Triangulation The objective of the presented work is to elaborate numerical algorithms for an automatic mesh generator for 2D finite element programs. The generator developed by the Authors is based on the well-known Delaunay triangulation [1]. The triangular mesh is generated using the description of the domain boundary only. Mesh density and shape of elements are controlled by a suitable distribution of the boundary nodes as well as the way in which the internal nodes are introduced. Estimation of the quality of generated meshes for different methods of internal nodes insertion and mesh quality improvement is presented.
B. Glut(1), K. Boryczko(1), T. Jurczyk(1), W. Alda(1), J. Kitowski(1,2)
(1) Institute of Computer Science AGH, Kraków, Poland
(2) ACC CYFRONET AGH, Kraków, Poland
15:15
CE6 Parallel Mesh Generation for Finite Element Modeling In this paper, a parallel mesh generation algorithm for the finite element modeling is proposed. The algorithm can be used on parallel machines and networks of heterogeneous workstations. The interial method is utilized to decompose a domain being discretized into a set of subdomains, and each of these subdomains is assigned to a separated processor. On each processor the corresponding subdomain is then meshed sequentially. The algorithm allows memory and time bottlenecks of a sequential mesh generation algorithm to be avoided.
Ł. Łaciński, R. Wyrzykowski
Technical University of Częstochowa, Poland
Session: Web Based Applications and Cellular Automata
Chair:
J. Nabrzyski
14:00
CF1 IERAPSI - Creating a Surgical Planning and Training System The latest development in advanced medical IT applications is the creation of complex integrated environments that support the physician in the operating theatre or in the preparation phase before entering the operation. In this paper we describe some aspects of a project that integrates advanced technologies such as 3D visualisation, virtual reality, and computerised simulation for pre-operative planning, and training of surgical procedures called 'Integrated Environment for the Rehearsal and Planning of Surgical Interventions' (IERAPSI). IERAPSI will particularly focus on the planning, simulation, and training of surgical procedures performed on the petrous bone, a common site with a complex anatomy. The resulting system will address frequently applied interventions, such as mastoidectomy, but equally very exacting ones, like cochlear implantation and acoustic neuroma. In a later stage, IERAPSI might also be used for other surgery specialities.
A. Emmen(1), L. Versweyveld(1), N.W. John(2)
(1) Genias Benelux, Almere, The Netherlands
(2) Manchester Visualisation Centre, Manchester Computing, University of Manchester, United Kingdom
14:15
CF2 VRML Based Sharing of 3D Models and Simulation Results in Global Collaborative Environment In the past engineers and designers, gathered around one drafting board, performed technical discussions based on 2D drawings. This paper describes an approach, which has been undertaken within the cross-border company in order to improve the productivity of multinational product development teams. The global collaborative environment, where VRML format as a main engineering information carrier has been presented. Several applications for VRML are shown and discussed.
D. Kruziewicz, T. Nowak, M. Florkowski, M. Fulczyk, ABB Corporate Research, Kraków, Poland
14:30
CF3 Highly Integrated Industrial Process Modeling & E-business Environments At international level, the e-business environments have become an actual imperative condition for the economic competition. In the context of the evolution of the global market, each economical society must re-estimate the impact of electronic channels concerning all its processes, such as: sale, purchase, manufacturing, distribution, communication and management. From this point of view e-business environments are now the winners.
P. Mitrea(1), D. Grigoras(2)
(1) Software ITC SA Cluj-Napoca, Romania
(2) Techincal University "Gh. Asachi" Iasi, Romania
14:45
CF4 Cellular Automata BIORemediation System We present the CABIOR system developed on SGI Onyx2 for an application of parallel computing to the simulation of the bioremediation interventions on polluted soils. The CABIOR system is devoted to the needs of the bioremediation user which generally is not skilled enough in modelling or in computer science, so it has been designed in order to hide as many computational technical details as possible and at the same time to provide advanced features to analyze and to test the model outcome.
M.C. Baracca, P. Ornelli, G.Clai, ENEA HPCN, Bologna, ITALY
15:00
CF5 River Flows Modelled by Cellular Automata Cellular automata (CA) are proposed in the paper for modelling and simulating rivers flow in a given landscape. The landscape is represented by a regular CA mesh. A single cell is defined by two attributes: altitude of the ground level and the amount of water. The water is exchanged between CA according to two different sets of rules assumed. Sample results generated on the SGI/Onyx system are also presented.
P. Topa, Institute of Computer Sciences, AGH, Poland
15:15
Application of Computer Simulations to Phase Transitions in the Ashkin-Teller Model Computer simulations have been used to analyse the critical behaviour of the Ashkin-Teller model: a) in the extended version in two dimensions and b) in the standard version in three dimensions. The simulations are of Monte Carlo type, based on the invariance of the ratio of the square of the second moment of the order parameter to its fourth moment in the critical region. The critical surface is calculated in regions where the continuous transitions are expected both for the ferro- and the antiferromagnetic regions. The continuity of the order parameter at the critical surface is verified by a finite size scaling analysis for both feromagnetic and antiferromagnetic parts of the phase diagram. Simulations have been performed on SGI Power Challenge XL supercomputers using the 64-bit random number generator.
CF6
G. Musiał, I. Dębski, G. Kamieniarz, Institute of Physics, A. Mickiewicz University, Poznań, Poland
Session: Numerical Linear Algebra
Chair: A. Emmen
15:45
CG1 Monte Carlo Method for Solution of Linear Algebraic Equations
with Chain Reduction and Optimization Using PLFG 		In this paper, we show a Monte Carlo method for solution of linear algebraic equations with chain reduction and optimization, coupled with PLFG, a parallel pseudo-random generator. PLFG, designed for MIMD architectures, is highly scalable and with the default parameters chosen, it provides an astronomical period of at least 2**29 (2**23209 - 1). Numerical experiment results show that Monte Carlo method with chain optimization and reduction gives much estimates of the solution vector.
Keywords: Monte Carlo method, Pseudo-random number generator, Lagged Fibonnaci Generator, Parallel computation, Solution of linear algebraic equations.
C. J.K. Tan(1), M.I.C. Villalba(2), V. Alexandrov(1)
(1) High Performance Computing Center, Univ. of Reading, United Kingdom
(2) Norkom Technologies Ltd., Dublin, Ireland
16:00
Recursive Approach in Sparse Matrix LU Factorization This paper describes the recursive LU factorization of sparse matrices.The recursive formulation of common linear algebra codes has been proven very successful in dense matrix computations. We present an extension of this idea to sparse matrices. Our experiments show that the recursive approach can be very competitive with leading tools for sparse martix factorization in terms of time, required storage space, and error estimation.
CG2
J. Dongarra, V. Eijkhout, P. Łuszczek, Dept. of Computer Science, University of Tennessee, USA
16:15
CG3 Effectiveness of Adaptively Left Preconditioned GMRES(m) Algorithm on Origin 2000 GMRES(m) method is one of the iterative algorithms for solving the linear systems of equations with a large, sparse, and nonsymmetric coefficient matrix. This restarted method is able to reduce both computational work and its memory requirement. However, using the restart, GMRES(m) method looses the information of some eigenvectors corresponding to the smallest eigenvalues, and its convergence may slow down or stall. In this paper, we compare the restarted GMRES(m, k, a, b) method [1] which determines left preconditioner adaptively coupling with IRA (Implicitly Restarted Arnoldi) process [2]. At last, we will give the numerical experiments on the parallel machine Origin 2000 and concluding remarks.
H. Suzuki(1) and T. Nodera(2)
(1) Graduate School of Science and Technology, Keio University, Japan
(2) Faculty of Science and Technology, Keio University, Japan
16:30
CG4 Modeling the Behaviour of Linear Algebra Algorithms: a Study with Jacobi Methods on Origin 2000 The modeling of the behaviour of linear algebra algorithms is very interesting in order to design linear algebra software for high performance computers. This modelization would enable us to predict the execution time of the routines depending on a number of parameters. There are two groups of parameters, in the first, there are the parameters whose values we can choose: number of processors, processors grid configuration, distribution of data in the system, block size; and in the second, we have the parameters that specify the characteristics of a target architecture: arithmetic cost and start-up and word-sending cost of a communication operation. In this way, linear algebra library could be designed where each routine takes the values of the parameters of the first group with which the optimum execution time is predicted, and solves the problem. So, this library could be employed by a non expert user to solve scientific or engineering problems, because the user does not need to determine the values of these arameters. The design methodology is analysed with one-sided block Jacobi methods to solve the symmetric eigenvalue problem. Variants for a logical ring and a logical rectangular mesh of processors are considered. An analytical model of the algorithm is developed, and the behaviour of the algorithm is analysed in an Origin 2000 with message-passing using MPI.
J. Cuenca(1), D. Giménez(2), J. González(1)
(1) Departamento de Ingeniería y Tecnología de Computadores, Universidad de Murcia, Spain
(2) Departamento de Informática, Lenguajes y Sistemas Informáticos, Universidad de Murcia, Spain
16:45
CG5 An Effective Algorithm for Generation of Closed Connected Linear Graphs An effective algorithm for generation of closed connected linear graphs in any dimension is proposed. It is an extension of earlier algorithms based on collapsing the unlinked vertices starting form polygons. Novel in our approach is that for each obtained graph we introduce new links between unlinked vertices and we decorate bonds with new vertices. In this way we obtain the more complex graphs in an early stage of the graph generation procedure. We have performed calculations on SGI Power Challenge XL supercomputers and the results for different generation methods are compared.
G. Musiał(1), J. Rogiers(2)
(1) Institute of Physics, A. Mickiewicz University, Poznań, Poland
(2) Institute for Theoretical Physics, K.U. Leuven, Belgium
Session: Miscellaneous Applications
Chair: tba
15:45
CH1 SGI Computers in the Polish Navy C4I System The article describes one of the numerous software tools developed at OBR CTM and used in the Polish Navy's C4I system based on SGI technology. The tool being in fact a set of graphical functionality supports all operators at Headquarters and aboard ships in their daily duties. Duties are concentrated on graphical presentation of present tactical situation and preparation of different graphical documents. The tool, being a fully integrated part with the Navy's system, provides very specialised features which can not be offered by other tools available on the market.
A. Kilian, OBR Centrum Techniki Morskiej, Gdynia, Poland
16:00
CH2 Parallel Processing of the Configuration Space of Robots In this article it is presented the design and implementation of algorithms for the calculation of the configuration space for a robot upon a distributed computing system. In these algorithms two levels of parallelism appear: for the basic level of operation of the proposed algorithm and for the level of the computational tool utilized -Fast Fourier Transform (FFT). The library complib.sgimath has been used. In the method employed to carry out the paralellization the master-slave philosophy is followed, using a message-passing library (MPI). Taking as case of study a robot that moves upon a plan, it is carried out a load balance and the influence of the granularity is analyzed or a master and various slave processes over a tetraprocessor SGI Origin 200. In this way, it is obtained a reduction of the calculation time, for the best solution, to a third part of the employed time in the implementation of the secuential algorithm, showing the line to follow for the significant reduction of the calculation times.
R. Theron, J. Blanco, B. Curto, V. Moreno, Dept. Computers and Automation, Univ. of Salamanca, Spain
16:15
CH3 STL Based Library for Evolution Programs In this paper, we present an object oriented library for evolution programs, developed in the Technical University of Częstochowa. This package is based on the SGI Standard Templates Library. It uses STL memory management as well as some generic algorithms. Presented library offers advantages of optimized kernel and flexible user interface. Its main features are illustrated by the example of optimizing De Jong test function.
J. Żola, R. Wyrzykowski, Institute of Mathematics and Computer Science,Technical University of Częstochowa, Poland
16:30
Free Vibration and Forced Vibration Analysis in Induction Motor In addition to the static forces resulting form the engine operation to produce the driving torque, also dynamic forces act upon the rotor of the electric motor act. These forces are due to the mechanical and electrical motor features. The mechanical forces result from the rotor unbalance allowances or damages to the elements of rotor bearings. The forces of the magneto-electric origin are due among the others to the change of the length of the airgap between the rotor and stator caused by their misalignment. The alternating forces generated in the rotor can be amplified when to the frequencies of the free vibration of the rotor-stator unit. Thus it is important to know the
distribution of the free vibration frequencies for the whole unit. To determine the risky areas of the motor operation it is necessary to perform the calculations of the free vibration frequencies for the
whole unit. In the model presented here the rotor shaft complete with core plates and the cage winding. Additionally the model includes also the housing parts complete with two covers, stator core plates and winding elements. The bearings placed in the stator covers were modelled with elastic elements having the appropriate elasticity. Values for the free vibration of such model were determined with finite elements method. The rotor geometry was modelled in the MSC/PATRAN system. To determine the eigen values ABAQUS v5.8 program was used. The calculations were performed in the Academy Computer Center in Cracow, Poland.
CH4
M. Noga(1), L. Gołębiowski(2), M. Gołębiowski(2), D. Mazur(2)
(1) ACC CYFRONET AGH, Kraków, Poland
(2) Rzeszów University of Technology, Poland
16:45
CH5 Visualization Environment for 3D Modelling of Numerical Simulation Results The paper is devoted to the complex process that begins by obtaining numerical data while solving a scientific problem and completes by the creation of the demo for illustrating the final solution in the graphical form. The represented visualized results were obtained after the simulation of the particle movement in the system of three bodies on the LaGrange surface, the simulation of mature convective clouds development and aspirin modelling. All calculations were carried out using supercomputers of CONVEX C-series, PARSYTEC systems and Cray J-90. Then the numerical data was converted using Geomview - special viewing program oriented for 3D visualization under IRIX 6.2 (SGI Octane workstation). It became possible thanks to the additional module written in C/C++ and aimed to convert data of special formats into the series of graphical images. As a result, 3D animation films were created.
A.V. Bogdanov, E.N. Stankova, E.V. Zudilova, COMMOD s.r.l. Roma, Italy