Parareal Acceleration of Matrix Multiplication
Toshiya Takami, Nishida Akira · Advances in parallel computing · 2012
The parareal-in-time algorithm has been used in time-domain parallelism of scientific problems described by ordinary differential equations and partial differential equations. In this paper, applicability of this algorithm to simple linear transformations is studied through convergence and speedup ratio. These properties are analyzed, in particular, for calculations of series of vectors defined by matrix multiplications. The actual speedup ratio is measured in a distributed parallel machine and compared to theoretical prediction. Finally, appropriate problem sizes for this scheme is presented based on these results.