Towards an Exascale Enabled Sparse Solver Repository
As we approach the exascale computing era, disruptive changes in the software landscape are required to tackle the challenges posed by manycore CPUs and accelerators. We discuss the development of a new ‘exascale enabled’ sparse solver repository (the ESS
- PDF / 386,905 Bytes
- 22 Pages / 439.36 x 666.15 pts Page_size
- 56 Downloads / 198 Views
Abstract As we approach the exascale computing era, disruptive changes in the software landscape are required to tackle the challenges posed by manycore CPUs and accelerators. We discuss the development of a new ‘exascale enabled’ sparse solver repository (the ESSR) that addresses these challenges—from fundamental design considerations and development processes to actual implementations of some prototypical iterative schemes for computing eigenvalues of sparse matrices. Key features of the ESSR include holistic performance engineering, tight integration between software layers and mechanisms to mitigate hardware failures.
1 Introduction It is widely accepted that the step from peta- to exascale is qualitatively different from previous advances in high performance computing and therefore poses urgent questions. Considering applications that need these vast computing resources, which algorithms expose such massive parallelism? What will the next generations of supercomputers look like, and how can we write sustainable yet efficient software
J. Thies () • M. Röhrig-Zöllner • A. Basermann Simulation and Software Technology, German Aerospace Center (DLR), Köln, Germany e-mail: [email protected] M. Galgon • B. Lang School of Mathematics and Natural Sciences, University of Wuppertal, Wuppertal, Germany A. Alvermann • A. Pieper • H. Fehske Institute of Physics, University of Greifswald, Greifswald, Germany M. Kreutzer • F. Shahzad • G. Hager • G. Wellein Erlangen Regional Computing Center, Friedrich-Alexander-University Erlangen-Nuremberg, Erlangen, Germany © Springer International Publishing Switzerland 2016 H.-J. Bungartz et al. (eds.), Software for Exascale Computing – SPPEXA 2013-2015, Lecture Notes in Computational Science and Engineering 113, DOI 10.1007/978-3-319-40528-5_13
295
296
J. Thies et al.
for them? The ESSEX project1 has developed the ‘Exascale enabled Sparse Solver Repository’ (ESSR) over the past three years, and in this paper we want to share our experiences and summarize our results in order to contribute to answering these questions. Besides reviewing the ESSEX project, the paper contributes a thorough presentation of a software architecture for iterative sparse solver libraries on heterogeneous supercomputers that overcomes some of the shortcomings of existing packages on the road to exascale. The applications we study come from quantum physics and material science, and are directly or indirectly related to solving the Schrödinger equation. The Hamiltonian of the systems studied can be represented as a (very) large and sparse matrix, and the numerical task is to solve sparse eigenvalue problems in various flavors. The software we develop is intended as a blueprint for other applications of sparse linear algebra. In the next few years, we expect no radical change in the architecture of supercomputers, so that a scaled up version of current petascale systems is used as target architecture for the ESSR. That is, a distributed memory cluster of (possibly heterogeneous) nodes. On the other hand, nod
Data Loading...
