A major feature of Computer Algebra, and more generally of non-numerical computations, is the dynamical and non-predictable behaviour of the executions. We then understand that statical analysis should imperatively be completed by dynamical analysis in order to reach the best distribution of the tasks among the processors. In this paper, we present a new load-balancing system for parallel architectures with great numbers of processors. Being well suited for Computer Algebra and based on the notion of granularity, it is original in the sense that it takes into account the tasks complexity as a consistent information in order to achieve efficiency.
Jean-Louis Roch, A. Vermeerbergen, Gilles Villard