banner page
Request for participation (.doc or .txt)
Request for financial assistance (.doc or .txt)
Numerical modelling of natural phenomena today requires an enormous amount of computer power, more than was even conceivable a few years ago, as the speed of computer processors increases by three orders of magnitude every 10 years. This power increase leads researchers to invent new numerical methods which take advantage of new computer architecture. In particular, by assigning several processors to one task, both the computing power and the memory capacity are enhanced. This leads to the search for new parallel algorithms.
A special effort has gone into the development
of linear algebra procedures since, in many numerical simulations,
they make up the largest part of the computation. As a result
of this special effort, a particular library - the LAPACK library
- has become the standard library for the manipulation of matrices
of intermediate size. When very large matrices (of orders ranging
from 104 to 106) are to be manipulated, methods which do not transform
or only slightly transform the matrices are better suited. By
the end of the last decade, research on these topics had produced
a quite clear picture; however, there remain many situations in
which direct methods are still not applicable and for which some
iterative methods do not converge or converge too slowly. A store
of remedies can then be used to attempt to bring a partial response to
these remaining difficulties.
This set of courses will describe the state-of-the-art for programming procedures for large problems in linear algebra.
Some of the courses are devoted to the definition of currently used methods. Special attention is given to the race for efficiency which opposes iterative and direct methods for solving large linear systems.
Other courses focus on the parallelization of the methods and on their implementation on networks of workstations or on multiprocessors.
By analysing all the steps of the numerical simulation, from modelling to computer implementation, it is shown, using a particular application - underground water flow - how to build efficient software which can exploit local or exterior computing resources.
Patrick Amestoy, ENSEEIHT,
Toulouse
Jocelyne Erhel, INRIA, Rennes
Jérôme Jaffré, INRIA,
Rocquencourt
Mohamed Jemni, Université de Tunis
Hugues Leroy, INRIA Rennes
Gérard Meurant, CEA/DIF, Bruyères
le Châtel
Bernard Philippe, ENIT/LAMSIN, Tunis
Jean Roberts, INRIA, Rocquencourt
Ahmed Sameh, Purdue University
N. Hariga (ENAT, Tunis),
D. Mezher (ESIB, Beyrouth),
R. Touihri (IPEIM, Monastir)
Bernard Philippe (coordinator)
Mohamed Abdelwahed, Lamia
Belaid Jaafar, Henda El Fekih, Moncef Mahjoub
The courses will be given in French and/or in English, depending on the composition of the audience.
Each course is one-week long. It is followed
up by lab exercises which are organized by members of LAMSIN.
Lab sequences will conclude with the installation and use of
up-to-date software.
PARALLELIZATION
PARA 1 : Introduction to Parallel Algorithms and Architectures. M. Jemni
PARA 2 : Hands on Grid computing, distributed and parallel programming. H. Leroy
PARA 3 : Parallelism in Numerical Linear Algebra. A. Sameh
METH 1 : Iterative methods for solving linear systems. G. Meurant
METH 2 : Linear algebra and sparse direct methods. P. Amestoy
METH 3 : Eigenvalue problem solvers. B. Philippe
APPL 1 : Cell-centered discretization methods for 2nd order elliptic problems. J. Jaffré and J. Roberts
APPLI 2 : Numerical models and grid computing applied to hydrogeology problems. J. Erhel
Weeks |
Lecturers |
Contents |
|
2nd to 7th of February
|
M. Jemni
|
PARA 1 (lecture and lab)
|
|
9th to 14th of February
|
H. Leroy
|
PARA 2 (lecture and lab)
|
|
16th to 21st of February
|
N. Hariga - B. Philippe
|
Sparse matrices (lab)
|
|
23rd to 28th of February
|
J. Jaffré - J. Roberts
|
APPL 1 (lecture)
|
|
1st to 6th of March
|
A. Sameh
|
PARA 3 (lecture)
|
|
8th to 13th of March
|
B. Philippe
|
PARA 3 (lab)
|
|
15th to 20th of March
|
G. Meurant
|
METH 1 (lecture)
|
|
22nd to 27th of March
|
BREAK
|
|
|
29th of March to 3rd of April
|
J. Erhel - N. Hariga
|
APPLI 2 (lecture and lab)
|
|
5th to 10th of April
|
R. Touihri
|
METH 1 (lab)
|
|
12th to 17th of April
|
P. Amestoy
|
METH 2 (lecture)
|
|
19th to 24th of April
|
N. Hariga - D. Mezher
|
METH 2 (lab)
|
|
26th to 30th of April
|
B. Philippe - R. Touihri
|
METH 3 (lecture and lab)
|
This program is open to all researchers and engineers who would like to acquire the skills necessary for programming numerical models in a way that best uses the capacities of the computer. It is especially well-suited to PhD students.
The necessary background in mathematics
is the bachelor's level. In computer science, the attendees should
know how to program in at least one of the languages: C, FORTRAN, Matlab,
or SCILAB.
There is no registration fee for the course.
Some fellowships for reimbursing living
and travel expenses will be granted to PhD students from African countries
who are sponsored by their universities.
Bernard Philippe
LAMSIN - ENIT
1002 TUNIS BELVEDERE
TUNISIA
phone:+216 71 874 700 ext : 555