About me

Xavier Besseron
Computer Science and Communications research unit
University of Luxembourg
Campus Kirchberg, Office E-007
Email: <xavier besseron (a) uni lu>

My name is Xavier Besseron and I am a postdoc researcher at the University of Luxembourg.

I have a PhD degree in Computer Science: more information. I also worked as a postdoc researcher at the Ohio State University.


Research

Research interests

Research Associate at the University of Luxembourg

I am Research Associate in the Research Unit in Engineering Science (RUES) of the University of Luxembourg. I am part of the Luxembourg XDEM Research Group which is led by Prof. Bernhard Peters. I work on the parallelization and design of the eXtedned Discrete Element Method (XDEM) software. I also ensure the software quality by maintaining the infrastructure for automatic testing and continuous integration.

I also collaborate actively with the Parallel Computing & Optimisation Group (PCOG) which is led by Prof. Pascal Bouvry from Computer Science and Communications research unit. In this area, my research is mainly focused on Fault-Tolerance, Energy efficiency and HPC execution on Distributed Platforms like Grids, Clusters and Cloud Computing environments.

Postdoc Researcher at the Ohio State University

I was a postdoc researcher at the Ohio State University from October 2010 to September 2011. I worked in the Network Based Computing Lab with is led by Prof. DK Panda. My research mainly focused on Fault-Tolerance for MVAPICH2 software.

PhD in Computer Science

I did my PhD at Laboratoire d'Informatique de Grenoble (LIG) -- INRIA in the MOAIS project. It deals with Fault Tolerance and Dynamic Reconfiguration for High Performance Computing. My supervisors were Thierry Gautier and Denis Trystram. I defended my PhD on April 2010.

I worked on the Kaapi middleware and I ran my experiments on the Grid'5000 platform. I have also participated in the IVth and Vth Grid Plugtests.

Title: Fault Tolerance and Dynamic Reconfiguration for Large Scale Distributed Applications

Abstract:

This work deals with high performance computing on large scale platforms such as computing grids. Computing grids are characterized by (1) frequent changes in execution context and, especially, by (2) a high failure probability caused by the large number of components. Running an application efficiently in such an environment requires to consider these parameters.

Our research work is based on the abstract representation of the application as a data flow graph from the parallel and distributed programming model Athapascan/Kaapi. This abstract representation is used to provide solutions for (1) dynamic reconfiguration and (2) fault tolerance issues.

These contributions are evaluated through the Kaapi and X-Kaapi software on the Grid'5000 computing platform.

Keywords: Parallel computing, Grid computing, Dynamic adaptation and reconfiguration, Fault tolerance, Data flow graph.

Publications

You can find a list of my publications here.


Experience

Software development

DPM

The Discrete Particle Method (DPM) is an advanced numerical simulation tool which deals with both motion and chemical conversion of particulate material such as coal or biomass in furnaces.

My work is to propose an enhanced design for high performance parallel execution of the DPM code. It is an extension of the previous work on Domain Decomposition which showed the benefit of parallel execution for the Dynamics module. The new design will allow parallel and distributed execution of coupled simulation modules of DPM. The targeted execution platforms are clusters of multiprocessor computers.

The primary goal of this enhanced design is to provide:

DPM Homepage

MVAPICH2

MVAPICH2 is a high performance MPI-2 implementation for InfiniBand, 10GigE/iWARP and RDMAoE based on MPICH2.

My main contributions to the MVAPICH2 project are

MVAPICH2 Homepage

FTB support in MVAPICH2

Kaapi

Kaapi is a C++ library that allows to execute multithreaded computation with data flow synchronization between threads. The library is able to schedule fine/medium size grain program on distributed machine. The data flow graph is dynamic (unfold at runtime). Target architectures are clusters of multiprocessor machines.

My main contributions to Kaapi are

Kaapi Homepage

Kaapi Forge

Grid user: Grid'5000

Grid'5000 is the French national grid for research purpose. It counts more than 5000 cores distributed over 9 sites.

Grid'5000 Homepage

During my PhD, I ran most of my experiments for Kaapi on the Grid'5000 platform.

Here is my Grid'5000 User report

Grid Plugtests Participant

I have participated in the IVth and Vth Grid Plugtests contests. These contests allowed to show the great performances of the Kaapi middleware.

More information

Teaching

From October 2006 to September 2009, I was a teaching assistant for the Université Joseph Fourier.

I taught Algorithmic and C programming under UNIX environment in DLST to undergraduate students.