Publicado em: 11/12/2009

Palestra com Prof. Vincent Danjean, de Grenoble, dia 11/12.

O professor Vincent Danjean, da Université Jospeh Fourier de Grenoble, está visitando o INF de 9 a 18 de dezembro, no âmbito do projeto CAPES/COFECUB coordenado pelo Prof. Philippe Navaux. No dia 11, sexta-feira, ele ministra a palestra “Scheduling Parallel Program with Work Stealing” às 14h, na sala 107 – prédio 43425.

O Prof. Danjean ensina no curso Polytech’Grenoble, de onde o INF já recebeu vários alunos de graduação em intercâmbio, inclusive para dupla diplomação. Sua área de pesquisa é a programação paralela, sendo ele membro do projeto INRIA MOAIS, e um dos principais desenvolvedores da linguagem paralela Kaapi.

Title: Scheduling Parallel Program with Work Stealing

Abstract: Nowadays, parallel programming is required to be able to archive lots of kind of computation (simulations, visualization of large data, …). To write such kind of parallel programs, several paradigm exists such as MPI and threads, OpenMP, … In this talk, we will focus on parallel programming with work stealing. With this model, we are able to prove optimal scheduling for some class of application.
Our team has developed a middleware, KAAPI. It allows to quickly write a parallel application based on work stealing. We will see how it works and how it has been optimized to minimize the overhead it induces on the program. Results with the STL algorithms are very interesting and exhibits good performances even with very small tasks.

CV: I defended my thesis in 2005 at the École normal supérieur de Lyon. I was working on user thread scheduling with respect to high speed communications with Raymond Namyst, my advisor. Since, I’ve been hired in the MOAIS INRIA research team at Grenoble, France. MOAIS in part of the LIG (Laboratoire d’Informatique de Grenoble/Grenoble computer science laboratory). Here, I’m working on parallel programming with work stealing on grids, multiprocessors and MPSoC. With a phd student, I’m also developing cache-oblivious and parallel algorithms for scientific visualization.