An Efficient Distributed Randomized Solver With Application to Large Dense Linear Systems
Source: University of Tehran
Randomized algorithms are gaining ground in high performance computing applications as they have the potential to outperform deterministic methods, while still providing accurate results. In this paper, the authors propose a randomized algorithm for distributed multicore architectures to efficiently solve large dense symmetric indefinite linear systems that are encountered, for instance, in parameter estimation problems or electromagnetism simulations. This solver combines an efficient implementation of a multiplicative preconditioning based on recursive random matrices, with a runtime (DAGuE) that automatically adjusts data structures, data mappings, and the scheduling as systems scale up.