Professional Experience

After one year at the University of Paris Orsay as assistant of research and teaching, I was nearly one year at the University of Houston Texas on the topic of OpenMP, language dedicated to manage parallel programmings according to a thread model and whose subjacent languages are FORTRAN and C++.

My work within the team of Mrs. Barbara Chapman consisted in finding methods to improve the performances on servers whose architecture is NUMA (Not Uniform Memory Access), i.e. whose access to the memory is not uniform according to the processor which tries to reach one zone of memory.

Among the servers now known respecting this NUMA paradigm, one can quote the servers containing Itanium. On the other hand, the servers based on processors like Sparc or Power are not NUMA (the access time to the memory is almost the same whatever the processor and the reached memory).

We again obtained performances known as supra-linear on standard mathematical problems (LU, Cholesky). (Extensions d’OpenMP pour les Architectures à Hiérarchie Mémoire Multiple, Achieving Performance under OpenMP on ccNUMA and Software Distributed Shared Memory Systems)