PERFORMANCE OF PARALLEL COMPUTATIONS OF TRIANGULAR STRUCTURE

By virtue of their inherent parallelism, matrix algorithms such as those for solving large systems of linear equations are of major importance in parallel computation. We are interested here in evaluating the speed-up obtained during the parallel execution of algorithms, whose task graph features a triangular structure. The mean execution time under probabilistic assumptions is computed for two different parallel execution schemes depending on the scheduling policy adopted Our modelling approach involves the development of an approximation technique for studying loosely coupled Markovian processes.