[Metalab] Please help me translate my thesis topic
Maciej Pasternacki
maciej at pasternacki.net
Fri Apr 3 17:29:21 CEST 2009
Hello,
I'm new here, so I'll introduce myself first. I'm a software (Lisp/
Python) geek that has just moved to Vienna, has been a few times at
the Lab, and at the moment finishes his Master's thesis. My problem
is that I'm still learning German and I'm glad when I can read a
children book, and my university wants me to give them my thesis topic
in foreign language if I want them to give me the diploma in that
language. So, I need to give them my topic in German. Would the
Lazylab help me with this?
My topic in English is "Hybrid parallelization of the UNRES molecular
dynamics force field algorithm" (original Polish, if that helps
anyone: Hybrydowa paralelizacja algorytmu pola siĆowego dynamiki
molekularnej UNRES).
UNRES (name stands for UNited RESidues) is program that tries to
predict structure (3D shape) of a protein particle, given the
protein's aminoacid sequence (protein structure prediction problem).
It employs methods of molecular dynamics (MD): physical simulation of
how the unfolded particle behaves in the solution (simulation of
various attraction and repulsion forces in discrete time steps). The
force field is the central part of algorithm: it is field function
that for every point in particle's conformational space (space defined
by all possible shapes of the particle) assigns particle's potential
and kinetic energy. This makes the whole simulation an optimizational
problem: we need to find the energy function's (force field's) global
minimum.
The whole program is parallelized to run on large clusters, using
message-passing paradigm (MPI standard), both coarse-grained (separate
simulations), and fine-grained (splitting single simulation onto
multiple cores). My work was to add fine-grained parallelization to
split single simulation ontu multiple cores, but using the threading
paradigm (OpenMP standard) and compare this with message-passing fine
grained parallelization. Hybrid means using both message-passing (for
coarse-grained parallelism) and threads (for fine-grained parallelism).
Within my work, I parallelize only the most time-consuming parts of
the force field functions, hence the "force field algorithm" part of
title.
Thanks in advance for any help,
regards, :)
Maciek.
--
Maciej Pasternacki -><- http://www.pasternacki.net/ -><- http://www.3ofcoins.net/
More information about the Metalab
mailing list