|
|
Large-scale geosimulations using GEM and trilions
Blaheta, Radim ; Jakl, Ondřej ; Hrtus, Rostislav ; Kolcun, Alexej ; Kohut, Roman ; Starý, Jiří ; Turan, E.
The project deals with efficient solution of singular symmetric positive semidefinite problems. Our motivation arises from the \nneed to solve special problems of geotechnics, e.g. to perform upscaling analysis of geocomposites. In that and other applications we have to solve boundary problems with pure Neumann boundary conditions. We show that the stabilised PCG method with various preconditioners is a good choice for systems resulting from the numerical solution of Neumann problems, or more generally problems with a known small dimensional null space. The work was also taken as an opportunity to compare parallel implementations of the orresponding solvers, namely implementations in the in-house finite element software GEM and implementations employing components of the general Trilinos library. The studies show that the solvers based on GEM are highly competitive with its recognised counterpart.
|
|
|
Iterative solution of singular systems with applications
Blaheta, Radim ; Jakl, Ondřej ; Starý, Jiří ; Turan, E.
This contribution concerns the iterative solution of singular systems which arise in many applications. Let us mention the following: - solution of PDE problems with pure Neumann boundary conditions (which is our main aim), see [7], [8], [20]. Such problems have a specific role in numerical upscaling, see [6], - solution of Neumann type subproblems in domain decomposition techniques as FETI, Neumann-Neumann, BDDC methods, see [22], [16], - analysis of Markov chain problems, computation of stochastic vector, see e.g. [18], [19], - computer tomography [15], [14] and inverse problems [4], [21].
Fulltext: content.csg -  PDF
Plný tet: UGN_0426825 - PDF
|
|
|
High performance computing in micromechanics
Blaheta, Radim ; Hrtus, Rostislav ; Jakl, Ondřej ; Starý, Jiří
By micromechanics we understand analysis of the macroscale response of materials through investigation of processes in their microstructure. Here by the macroscale, we mean the scale of applications, where we solve engineering problems involving materials like different metals and composites in aircraft design or rocks and concrete in a dam construction. Different applications are characterized by different characteristic size. At macroscale the materials mostly look as homogeneous or they are idealized as homogeneous or piecewise homogeneous. A substantial heterogeneity is hidden and appears only after more detailed zooming view into the material.
Fulltext: content.csg - PDF
Plný tet: UGN_0426823 - PDF
|
|
|
Our Blue Gene Experience
Jakl, Ondřej ; Starý, Jiří
The paper describes our precious short-time opportunity to make some hands-on experience with the Bulgarian IBM Blue Gene/P system, where we tested the parallel scalability of our domain decomposition finite element solver on a greater number of processors and completed experiments with microstructure modelling of geocomposite materials. The tests showed that due to its relatively low computation/communication performance ratio this specific architecture may need optimizations of parallel codes to take advantage of its performance potential.
|
|
|
Využití DCT/DCE v paralelních výpočtech rotorů
Jakl, Ondřej ; Musil, Tomáš
Four years ago, The MathWorks, the producer of Matlab, fulfilled the wishes of many users and released the first version of the DCT/DCE tools enhancing Matlab’s capabilities towards parallel processing, to take advantage of the modern multiprocessor/multicore computer architectures. We followed DCT/DCE’s development, investigated this computing environment from the point of view of the “classical” parallel programming and took advantage of it in a set of Matlab codes for modelling of nonlinear dynamics of rotors. This paper summarizes our (throughout positive) experience.
|
|
|
Úvod do paralelního Matlabu
Jakl, Ondřej ; Musil, Tomáš
The contribution provides an introduction to the DCT/DCE parallel computing environment which has recently become part of the Matlab software collection. An application from the field of nonlinear dynamics of rotors is presented and selected parallel constructs are introduced. Practical experiments with this application show very good parallel performance and scalability on a local cluster of 8 processors.
|
|
|
Matlab paralelně
Jakl, Ondřej ; Musil, Tomáš
Matlab is a popular software widely used for technical and scientific computing. Quite recently, its producer The MathWorks released an add-on toolbox enhancing Matlab’s capabilities towards parallel processing. We investigated this environment and developed some parallel Matlab codes for modelling of nonlinear dynamics of rotors. This paper summarizes our (throughout positive) experience.
|
|
|
Parallel algorithm of modified Newmark's method in dynamics of rotors
Musil, Tomáš ; Jakl, Ondřej
Parallel algorithm of modified Newmark's method is used in dynamics of rotors supported by finite length journal bearings. The algorithm, realized in the MATLAB parallel computing environment (DCT/DCE), was tested on two different types of parallel computers – a Beowulf cluster and symmetric multiprocessor. Tests revealed good scalability and efficiency of proposed algorithm.
|
|
|
25 let vědeckých výpočtů na Ústavu geoniky
Jakl, Ondřej
The paper summarizes the role of computers in the twenty five years since the establishment of the Institute of Geonics Cz.Acad.Sci. It recalls the importance of computer systems and computer science in general, especially for mathematical modelling, one of the most important research fields of the institute. In the course of time computers, networks and Internet are unavoidable for each of the fields of study of the institute. The contribution surveys the main stages in this development, from the point of view of demanding computations in particular, and mentions various computers and hardware characteristic for each stage.
|
|
|
Poslední trendy v paralelním programování
Jakl, Ondřej
The paper gives a survey on current approaches and trends in programming parallel systems, paying special attention to scientific and high-performance computing. The mainstream tools (libraries, languages, products) for systems with shared and distributed memory are characterized and some novelties like parallel Matlab are introduced.
|
guest ::
RSS feed.