Applied numerical linear algebra
Applied numerical linear algebra
Accelerating configuration interaction calculations for nuclear structure
Proceedings of the 2008 ACM/IEEE conference on Supercomputing
Generic topology mapping strategies for large-scale parallel architectures
Proceedings of the international conference on Supercomputing
Exploring the future of out-of-core computing with compute-local non-volatile memory
SC '13 Proceedings of the International Conference on High Performance Computing, Networking, Storage and Analysis
Hi-index | 0.00 |
Obtaining highly accurate predictions for properties of light atomic nuclei using the Configuration Interaction (CI) approach requires computing the lowest eigenvalues and associated eigenvectors of a large many-body nuclear Hamiltonian matrix, $\hat{H}$. Since $\hat{H}$ is a large sparse matrix, a parallel iterative eigensolver designed for multi-core clusters is used. Due to the extremely large size of $\hat{H}$, thousands of compute nodes are required. Communication overhead may hinder the scalability of the eigensolver at such scales. In this paper, we discuss how to reduce such overhead. In particular, we quantitatively show that topology-aware mapping of computational tasks to physical processors on large-scale multi-core clusters may have a significant impact on efficiency. For typical large-scale eigenvalue calculations, we obtain up to a factor of 2.5 improvement in overall performance by using a topology-aware mapping.