Full HTML for

Basic foilset Overview of NPAC Research in Computational Chemistry

Given by David Bernholdt at Visit by CEWES Team on Mar 25 1997. Foils prepared April 5 97
Outside Index Summary of Material

Use of computational simulations to better understand, interpret, and predict chemical phenomena.
Current work includes
  • Application of existing tools to interesting chemical problems
  • Development of new tools to expand the range of systems we can study
The desire to effectively utilize massively parallel computers drives much of the development work

Table of Contents for full HTML of Overview of NPAC Research in Computational Chemistry

 Denote Foils where Image Critical
Denote Foils where Image has important information
Denote Foils where HTML is sufficient
1 Overview of NPAC Research in Computational Chemistry
2 Statement of Interests
3 Projects Underway or Anticipated
4 Major Collaboration: Pacific Northwest National Laboratory
5 NWChem Background
6 General HPC Tools Underlying NWChem
7 Global Array Toolkit Background
8 Global Arrays
9 Global Array Operations
10 The Hartree-Fock (SCF) Method
11 Hartree-Fock using Global Array Programming Model
12 Fock Matrix Construction – Parallel Performance (Intel Delta)
13 DFT Parallel Scaling on a Zeolite Fragment (Si28O67H30)
14 Availability of Tools

Outside Index Summary of Material

HTML version of Basic Foils prepared April 5 97

Foil 1 Overview of NPAC Research in Computational Chemistry

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
David E. Bernholdt
NPAC / Syracuse University
HTML version of Basic Foils prepared April 5 97

Foil 2 Statement of Interests

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
Use of computational simulations to better understand, interpret, and predict chemical phenomena.
Current work includes
  • Application of existing tools to interesting chemical problems
  • Development of new tools to expand the range of systems we can study
The desire to effectively utilize massively parallel computers drives much of the development work
HTML version of Basic Foils prepared April 5 97

Foil 3 Projects Underway or Anticipated

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
Parallelization of MOPAC
Development of novel approaches to the MP2 problem suitable to large-scale calculations on MPPs
Parallel I/O in the SCF method (NWU, UIUC)
Parallel CCSD energy and gradients (PNNL)
DOE Grand Challenge "Computational Chemistry of Nuclear Waste Characterization and Processing: Relativistic Quantum Chemistry of Actinides" (PNNL, ANL, Eloret Inst., OSU, Duke, UT-Austin)
Various applications (PNNL, SU Chemistry)
HTML version of Basic Foils prepared April 5 97

Foil 4 Major Collaboration: Pacific Northwest National Laboratory

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
DOE laboratory, in Richland, WA (Hanford Reservation)
DEB is Affiliate Staff Scientist at PNNL
Environmental Molecular Sciences Laboratory
  • Home of most powerful IBM SP installation anywhere (at present)
    • 472 nodes (432 compute), 200 GFLOPs, 62 GB memory, 1.9 TB disk, 5.5 GB/s aggregate sustainable bandwidth to disk
  • High-Performance Computational Chemistry Group leads development of NWChem parallel computational chemistry package.
HTML version of Basic Foils prepared April 5 97

Foil 5 NWChem Background

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
Most extensive "designed parallel" computational chemistry code available
Includes molecular dynamics (MD, MC), semiempirical, ab initio (SCF, MC-SCF, MP2, MR-CI, CCSD), and DFT
Works efficiently on both distributed- and shared- memory architectures
Supported on: Linux laptops, most unix workstations, workstation networks, SGI SMP systems, KSR, Intel Paragon, IBM SP, Cray T3x
Focus on scalability of algorithms, code
HTML version of Basic Foils prepared April 5 97

Foil 6 General HPC Tools Underlying NWChem

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
Global Array Toolkit
  • Portable parallel programming model, similar to HPF
ChemIO parallel I/O library
  • Disk Resident Arrays (extension of Global Arrays to disk)
  • Exclusive Access Files (node-private files)
  • Shared Files (under development)
PeIGS parallel eigensolver
  • Fastest eigensolver currently available, sequential or parallel
  • Guarantees strong orthogonality across processors for degenerate systems (other parallel eigensolvers don't)
HTML version of Basic Foils prepared April 5 97

Foil 7 Global Array Toolkit Background

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
Wanted a standard, portable, vendor-supported parallel programming model
HPF-1 standard had not yet been finalized, existing compilers incomplete and offered inadequate performance
Chemistry applications require capabilities not in HPF-1 (task parallelism)
Solution: Develop GA Toolkit internally, with intention to transition to a later version of HPF
HTML version of Basic Foils prepared April 5 97

Foil 8 Global Arrays

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
"Shared-memory" programming interface for both shared- and distributed-memory computers
Processes can asynchronously access logical blocks of physically distributed matrices, without need for explicit cooperation by other processes.
Exposes the Non-Uniform Memory Access (NUMA) characteristics of the high performance computers
Designed to complement rather than replace the message-passing programming model.
HTML version of Basic Foils prepared April 5 97

Foil 9 Global Array Operations

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
Collective primitive operations
  • create a distributed array, controlling alignment and distribution
  • duplicate an array with properties of another existing array
  • destroy an array
  • synchronize
MIMD primitive operations
  • fetch, store and accumulate into a rectangular patch of global array
  • gather and scatter
  • atomic read and increment
  • efficient access to local elements of global array (by a pointer)
BLAS-like data-parallel operations
  • vector operations including: dot product, scale, add
  • matrix operations including: symmetrize, transpose, multiplication
Interface to third-party software
  • ScaLAPACK, PeIGS, SUMMA
HTML version of Basic Foils prepared April 5 97

Foil 10 The Hartree-Fock (SCF) Method

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
Each electron sees average field of other electrons
Iterative procedure to achieve a self-consistent field
Result is the familiar molecular orbital (MO) picture
Core of computation: construction of Fock matrix:
Fmn¬ Dls {2(mn|ls)-(ml|ns)}
(Gaussian) basis functions {cm} used to expand wave function
Integrals (mn|ls) are Coulomb repulsions of electrons in specific basis functions
HTML version of Basic Foils prepared April 5 97

Foil 11 Hartree-Fock using Global Array Programming Model

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
Subroutine library provides user with appearance of shared memory regardless of underlying platform
User writes program to account for remote memory access time
HTML version of Basic Foils prepared April 5 97

Foil 12 Fock Matrix Construction – Parallel Performance (Intel Delta)

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
HTML version of Basic Foils prepared April 5 97

Foil 13 DFT Parallel Scaling on a Zeolite Fragment (Si28O67H30)

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
HTML version of Basic Foils prepared April 5 97

Foil 14 Availability of Tools

From Overview of NPAC Research in Computational Chemistry Visit by CEWES Team -- Mar 25 1997. *
Full HTML Index
Global Array Toolkit
  • Includes GA, MA (memory allocator), TCGMSG (message passing), TCGMSG-MPI (TCGMSG--&#062 MPI conversion)
  • Public domain
  • ftp://ftp.pnl.gov/pub/permanent/global
PeIGS
  • Release 3.0 planned for May
  • Public domain
  • E-mail George Fann &#060gi_fann@pnl.gov&#062
ChemIO
  • Under development, public domain release planned
  • E-mail Jarek Nieplocha &#060j_nieplocha@pnl.gov&#062
© Northeast Parallel Architectures Center, Syracuse University, npac@npac.syr.edu

If you have any comments about this server, send e-mail to webmaster@npac.syr.edu.

Page produced by wwwfoil on Thu Aug 14 1997