Masha Sosonkina

pARMS: a parallel version of the algebraic recursive multilevel solver

A parallel version of the algebraic recursive multilevel solver (ARMS) is developed for distributed computing environments. The method adopts the general framework of distributed sparse matrices and relies on solving the resulting distributed Schur complement system. Numerical experiments are presented which compare these approaches on regularly and irregularly structured problems. Copyright

Accelerating configuration interaction calculations for nuclear structure

One of the emerging computational approaches in nuclear physics is the configuration interaction (CI) method for solving the many-body nuclear Hamiltonian in a sufficiently large single-particle basis space to obtain exact answers - either directly or by extrapolation. The lowest eigenvalues and corresponding eigenvectors for very large, sparse and unstructured nuclear Hamiltonian matrices are obtained and used to evaluate additional experimental quantities. These matrices pose a significant challenge to the design and implementation of efficient and scalable algorithms for obtaining solutions on massively parallel computer systems. In this paper, we describe the computational strategies employed in a state-of-the-art CI code MFDn (Many Fermion Dynamics - nuclear) as well as techniques we recently developed to enhance the computational efficiency of MFDn. We will demonstrate the current capability of MFDn and report the latest performance improvement we have achieved. We will also outline our future research directions.

Scaling of ab-initio nuclear physics calculations on multicore computer architectures

Abstract We discuss the scaling behavior of a state-of-the-art Configuration Interaction code for nuclear physics on modern multicore computer architectures. In the CI approach, the quantum many-body problem is expressed as a large sparse symmetric eigenvalue problem, of which the lowest eigenvalues and eigenvectors have to be computed. We compare the performance of the pure MPI version with the hybrid MPI/OpenMP code on Cray XT4 and XT5 platforms. For large core counts (typically 5,000 and above), the hybrid version is more efficient than pure MPI.

Algorithm 857: POLSYS_GLP—a parallel general linear product homotopy code for solving polynomial systems of equations

Globally convergent, probability-one homotopy methods have proven to be very effective for finding all the isolated solutions to polynomial systems of equations. After many years of development, homotopy path trackers based on probability-one homotopy methods are reliable and fast. Now, theoretical advances reducing the number of homotopy paths that must be tracked and handling singular solutions have made probability-one homotopy methods even more practical. POLSYS_GLP consists of Fortran 95 modules for finding all isolated solutions of a complex coefficient polynomial system of equations. The package is intended to be used on a distributed memory multiprocessor in conjunction with HOMPACK90 (Algorithm 777), and makes extensive use of Fortran 95-derived data types and MPI to support a general linear product (GLP) polynomial system structure. GLP structure is intermediate between the partitioned linear product structure used by POLSYS_PLP (Algorithm 801) and the BKK-based structure used by PHCPACK. The code requires a GLP structure as input, and although finding the optimal GLP structure is a difficult combinatorial problem, generally physical or engineering intuition about a problem yields a very good GLP structure. POLSYS_GLP employs a sophisticated power series end game for handling singular solutions, and provides support for problem definition both at a high level and via hand-crafted code. Different GLP structures and their corresponding Bezout numbers can be systematically explored before committing to root finding.

Performance Modeling and Analysis of a Massively Parallel Direct - Part 2

Modeling and analysis techniques are used to investigate the performance of a massively parallel version of DIRECT, a global search algorithm widely used in multidisciplinary design optimization applications. Several high-dimensional benchmark functions and real world problems are used to test the design effectiveness under various problem structures. In this second part of a two-part work, theoretical and experimental results are compared for two parallel clusters with different system scales and network connectivities. The first part studied performance sensitivity to important parameters for problem configurations and parallel schemes, using performance metrics such as memory usage, load balancing, and parallel efficiency. Here linear regression models are used to characterize two major overhead sources, interprocessor communication and processor idleness, and also applied to the isoefficiency functions in scalability analysis. For a variety of high-dimensional problems and large-scale systems, the massively parallel design has achieved reasonable performance. The results of the performance study provide guidance for efficient problem and scheme configuration. More importantly, the design considerations and analysis techniques generalize to the transformation of other global search algorithms into effective large-scale parallel optimization tools.

Ab initio nuclear structure ? the large sparse matrix eigenvalue problem

The structure and reactions of light nuclei represent fundamental and formidable challenges for microscopic theory based on realistic strong interaction potentials. Several ab initio methods have now emerged that provide nearly exact solutions for some nuclear properties. The ab initio no core shell model (NCSM) and the no core full configuration (NCFC) method, frame this quantum many-particle problem as a large sparse matrix eigenvalue problem where one evaluates the Hamiltonian matrix in a basis space consisting of many-fermion Slater determinants and then solves for a set of the lowest eigenvalues and their associated eigenvectors. The resulting eigenvectors are employed to evaluate a set of experimental quantities to test the underlying potential. For fundamental problems of interest, the matrix dimension often exceeds 1010 and the number of nonzero matrix elements may saturate available storage on present-day leadership class facilities. We survey recent results and advances in solving this large sparse matrix eigenvalue problem. We also outline the challenges that lie ahead for achieving further breakthroughs in fundamental nuclear theory using these ab initio approaches.

Design and implementation of a massively parallel version of DIRECT

Abstract This paper describes several massively parallel implementations for a global search algorithm DIRECT. Two parallel schemes take different approaches to address DIRECT’s design challenges imposed by memory requirements and data dependency. Three design aspects in topology, data structures, and task allocation are compared in detail. The goal is to analytically investigate the strengths and weaknesses of these parallel schemes, identify several key sources of inefficiency, and experimentally evaluate a number of improvements in the latest parallel DIRECT implementation. The performance studies demonstrate improved data structure efficiency and load balancing on a 2200 processor cluster.

Collective modes in light nuclei from first principles.

Results for ab initio no-core shell model calculations in a symmetry-adapted SU(3)-based coupling scheme demonstrate that collective modes in light nuclei emerge from first principles. The low-lying states of 6Li, 8Be, and 6He are shown to exhibit orderly patterns that favor spatial configurations with strong quadrupole deformation and complementary low intrinsic spin values, a picture that is consistent with the nuclear symplectic model. The results also suggest a pragmatic path forward to accommodate deformation-driven collective features in ab initio analyses when they dominate the nuclear landscape.

Computational nuclear quantum many-body problem: The UNEDF project

The UNEDF project was a large-scale collaborative effort that applied high-performance computing to the nuclear quantum many-body problem. The primary focus of the project was on constructing, validating, and applying an optimized nuclear energy density functional, which entailed a wide range of pioneering developments in microscopic nuclear structure and reactions, algorithms, high-performance computing, and uncertainty quantification. UNEDF demonstrated that close associations among nuclear physicists, mathematicians, and computer scientists can lead to novel physics outcomes built on algorithmic innovations and computational developments. This review showcases a wide range of UNEDF science results to illustrate this interplay.

Per-call energy saving strategies in all-to-all communications

With the increase in the peak performance of modern computing platforms, their energy consumption grows as well, which may lead to overwhelming operating costs and failure rates. Techniques, such as Dynamic Voltage and Frequency Scaling (called DVFS) and CPU Clock Modulation (called throttling) are often used to reduce the power consumption of the compute nodes. However, these techniques should be used judiciously during the application execution to avoid significant performance losses. In this work, two implementations of the all-to-all collective operations are studied as to their augmentation with energy saving strategies on the per-call basis. Experiments were performed on the OSU MPI benchmarks as well as on a few real-world problems from the CPMD and NAS suits, in which energy consumption was reduced by up to 10% and 15.7%, respectively, with little performance degradation.