UNIVERSITY OSLO
PHYSICS OF GEOLOGICAL PROCESSES
PGP EAN
Home
Publications
Resume

 
marcink at fys . uio . no

Professional experience

  1. Implementations of large resolution (hundreds of millions of mesh nodes) parallel numerical models on modern high-performance computing platforms (SMPs, GPUs, clusters/massively parallel supercomputers).

    • Numerical linear algebra, direct and iterative methods for the solution of large systems of linear equations

  1. Numerical solution of PDEs on different hardware platforms, using different methods

    • Large-resolution high-performance two- and three-dimensional Finite Elements solver for MATLAB (http://milamin.org/)

    • ADI/Fractional Steps methods on structured grids using shared memory computers and OpenMP / NUMAcc

    • FEM and unstructured adaptive meshes on high-performance supercomputers with thousands of CPU cores using MPI

    • Stencil computations and Geometric Multigrid for the Poisson's problem on modern GPUs using CUDA

  1. Various technical aspects of modeling

    • meshing/remeshing for unstructured computations

    • sparse matrix algebra

    • node renumbering, geometry (re)distribution

    • periodic boundary conditions

    • visualization of results using Paraview/VTK, MATLAB

  1. Software optimization and performance analysis on modern multi-core hardware.

    • memory bandwidth oriented optimizations for numerical codes

    • bottleneck identification (compute vs. memory bandwidth bounded codes)

  1. Seven years of work in collaboration with modelers from the Earth sciences. The challenges often addressed include

    • Complex geometries, heterogeneous materials, large resolution models

    • Different spatial and temporal scales of processes

    • Poisson's equation, Darcy flow, heat diffusion, transport, elastic wave propagation, incompressible Stokes flow

    • 2D and 3D data visualization and post-processing

  1. Ongoing projects

    MILAMIN 2 (http://milamin.org) is an integrated MATLAB environment for efficient 1D, 2D, 3D FEM modeling using unstructured grids. It includes optimized and parallelized MEX files for sparse matrix creation, sparse matrix-vector multiplication, quad/oct-tree generation and fast point in triangle location. The MEX files  can be freely downloaded from Sourceforge.

    MILAMIN supports a large variety of element types and provides a set of useful functions to manipulate the mesh, setup a numerical model, set the boundary conditions, solve the system of equations, and visualize the results. A large number of example models of different complexity are provided. The MILAMIN website provides benchmarks, as well as a detailed technical discussion of various steps of a state-of-the-art FEM solver.

  1. Programing languages (in order of expertise)

    • C (expert, memory bandwidth optimization, parallel programing with MPI, OpenMP, pthreads)

    • MATLAB (expert, 'vectorization' techniques, parallel MEX files for multi-core SMP architectures)

    • shell scripts

    • C++

    • Pearl scripts

    • Scientific Python

  1. Non-academia / company experience

    • distributed client-server systems, CORBA

    • public-key cryptography and SSL, secure data and AES

    • implementation of hardware drivers for Linux

    • shell and Pearl scripting on Linux-based systems

    • system programing on Linux-based workstations and embedded systems

    • software release management system (Pearl)

    • automatic compilation and assembly of custom Linux distribution for embedded systems

Current Post

Researcher, Physics of Geological Processes (PGP), University of Oslo, www.fys.uio.no/pgp

Member of the Resource Allocation Committee (RFK) appointed by the Research Council of Norway.

XI 2010, Defended the Degree of PhD

Title: Efficient implementations of numerical models for geological applications on modern computer architectures.

Thesis defended in the Faculty of Mathematics and Natural Sciences, Department of Informatics, University of Oslo

Opponents: Jason Phipps-Morgan, Cornell University; Petter E. Bjørstad, Department of Informatics, University of Bergen; Galen Gisler, Physics of Geological Processes, University of Oslo.

2008-2010, PhD Project, University of Oslo, Norway

Physics of Geological Processes (PGP), in collaboration with SINTEF

Relevant courses: Reservoir Geophysics, Introduction to Mechanical Geomodeling, Research Methods in Computer Science

2005-2007, Assistant Lecturer, University of Oslo, Norway

Physics of Geological Processes (PGP)

2002-2005, Software engineer, Sentivision Polska

Software designer and developer, software release manager, Linux-based embedded systems for VOD (Video on Demand) and distributed Content Management Systems.

1997-2002, Higher Education, Degree of Master of Science

Poland, Wroclaw University of Technology, Department of Software Engineering. Specialization:

  • Supercomputing, concurrent/parallel computing

  • Cryptography

  • Artificial Intelligence

Title of the Master thesis: Automatic parallelization of Java programs for distributed environments

The results were published in 2002 by IEEE Computer Society, "Amelia. Integrated Environment for Distributed Processing".