Abstract FR:

Ln part 1 of this thesis we consider parallel computer pro­ grams whose structure is governed by certain stochastic laws. Such programs are modeled by a random graph structure. An exact mathematical expression is given for the distribution of computing time with an infinite number of processors. A simulation validates the analytical model results. Ln part 2, a new method is proposed in order to optimize systolic networks for certains matrix computation. Its advantage is the improvement of processors utilization through halving the size of matrices. This improvement is exploited in an implementation of the Parallel Shooting, numerical method to solve two point differential equations. Interconnection network organization as well as algorithm execution are presented.