Doctor in Computer Science of the University of Bordeaux I in 1999 with congratulations of the jury, I began my occupations by continuing the topic of my doctoral research.

 

My doctoral research related to scientific computations requiring high performances according to the parallel paradigm of programming and on data that are represented by sparse matrices (i.e. including a great number of null values). These computations were carried out in FORTRAN using libraries of computations and executions which I had written in C language according to the model of programming known as High Performance FORTRAN (HPF).

 

This model makes it possible to obtain a program running on a multi processors server in a very optimized way and able to reach near linear performances (for 1 processor if the execution time is X, for N processors the time execution will be approximately X/n).

 

With the assistance of my research colleagues of Université of Bordeaux I, I created two libraries within the framework of my thesis: the Trident library whose goal was to allow to take into account in an optimal way sparse matrices in HPF and the Columbo library whose goal was to analyze computations to be realized in order to schedule them according to an optimal order.

 

The union of these two libraries made it possible to reach performances known as supra-linear (for N processors the execution time is lower than X/n), performances comparable to a program written completely in C, program dedicated to only one case of mathematical problem, but whose performances were proven like most powerful during several years.

 

My research made it possible to obtain in a relatively easy way high performance of computations for many scientific problems on the basis of sparse matrix.