High Performance Computing and Computational Science at AHPCC

High Performance Computing and High Performance Computing and Computational Science at AHPCCComputational Science at AHPCC

Brian T. Smith

Professor, Department of Computer Science

Director, Albuquerque High Performance Computing Center (AHPCC)

High Performance Computing Education & Research CenterHigh Performance Computing Education & Research Center

UNM strategic center to initiate and focus activities in high performance computing technology, research, and education

Mission accomplished through two centers

Established in 1994 as a training and resource center for MHPCC; now a national supercomputing Center within the NSF National Computational Science Alliance, serving as an academic center of excellence for research and education in computational science.

Established in 1994 under the auspices of the DoD Modernization Program, through a Cooperative Agreement between the University of New Mexico and the Air Force Research Laboratory. Provides production computing cycles for DoD researchers.

High Performance Computing, Education & Research CenterHigh Performance Computing, Education & Research CenterEXECUTIVE DIRECTOR

CO-DIRECTOR

CO-DIRECTOR

ASSOCIATE DIRECTOR

Frank L. Gilfeather

Brian T. Smith

John S. Sobolewski

Ernest D. Herrera

Maui High Performance Computing CenterMaui High Performance Computing CenterDIRECTOR

ASSOCIATE DIRECTORS

Eugene Bal

Gary Jensen

Steve Karwoski

Margaret Lewis

Albuquerque High Performance Computing CenterAlbuquerque High Performance Computing CenterDIRECTOR

ASSOCIATE DIRECTORS

Brian T. Smith

Susan R. Atlas

Robert A. Ballance

Ernest D. Herrera

Both centers support a significant number of users in academia and government, particularly the DoD and NSF, and are key players in the national supercomputer community.

MHPCCMHPCC

One of the top 30 supercomputingcenters in the world

A DoD Shared Center—a node on the National Technology Grid

65 staff members Computing systems

699 node IBM SP 400 GFLOPS computing power 167 GB total memory 2.1 TB internal disk storage 1.3 external disk storage 20 TB mass storage Visualization laboratory

0ver 1,100 government, industry, and academic users

AHPCCAHPCC

Ranks in the top 5 US academic institutions in supercomputing power (effective 5/00)

A member of the NSF Alliance and a node on the National Technology Grid

60 associated faculty, staff, postdocs and students Computing systems

512 processor IBM PIII Linux Supercluster (5/00) 128 processor Alta PII Linux Supercluster 32 processor VA Linux PIII Cluster Vista Azul - advanced IBM hybrid system 8 node SGI Origin 2000 16 processor Alta PII Linux development cluster Visualization laboratory

0ver 500 academic, industry, and government users

Supercomputing Capabilities

LosLobos & Roadrunner SuperclustersLosLobos & Roadrunner Superclusters

Research Environment at the AHPCCResearch Environment at the AHPCC

38 Graduate Research Assistants

16 Associated Faculty (Physics & Astronomy, Chemistry, Biology, Mechanical Engineering, Computer Science, EECE)

6 Permanent Research Staff

6 Visiting Scientists, Postdoctoral Fellows

Undergraduate Workstudy Students; NSF REU

Research Facilities: Supercomputers, High Performance Clusters, Workstations, Workshop Area, Seminar Room and Access Grid Studio

Educational Programs: SEC Program, Workshops, AHPCC Seminar Series, Alliance Activities, Native American Outreach, NSF AMO Summer School, UNM Course Laboratories

Computer Systems ResearchComputer Systems Research

To anticipate, develop, deploy, and support

high-performance computing technology and systems

Superclusters Open computing tools Grid-Based Computing Visualization

Superclusters: Beyond BeowulfSuperclusters: Beyond Beowulf

System design and integration Off-the-shelf symmetric multiprocessor subsystems High-speed interconnects Terabyte hierarchical mass storage systems

Research AreasResearch Areas Networking– Portals Hybrid (SMP) programming models Cluster Management– Maui Scheduler, PBS Condor high-throughput computing

Grid-Based Computing: Grid-Based Computing: Sharing Resources Across the MatrixSharing Resources Across the Matrix

Computational Grid: People to Machines, Machines to MachinesComputational Grid: People to Machines, Machines to Machines Globus Virtual Machine Room (VMR) Wireless networking

Access Grid: People to People and MachinesAccess Grid: People to People and Machines Telemedicine Visualization Human Factors Production Studio Deployment

Education & TrainingEducation & Training

TOUCH Telehealth Virtual CollaboratoryTOUCH Telehealth Virtual CollaboratoryDr. Dale Alverson (UNM), Dr. Richard Friedman (UH)Dr. Dale Alverson (UNM), Dr. Richard Friedman (UH)

• Access Grid multi-group Internet video conferencing for distance education

• Virtual Reality training environment

• 3D image/model manipulation and simulation environment using large, remote datasets

• Problem-based learning

Figure: A user and their “avatar” in the BioSIMMER environment (brain injury patient).

Scientific Visualization & Computational EnvironmentsScientific Visualization & Computational Environments

Visualization Laboratory – Homunculus Project “Flatland” Virtual Reality Environment Vista Azul Scalable Graphics Engine – parallel rendering CoMeT Computational Mechanics Toolkit Scientific Visualization Research

Science and Engineering ResearchScience and Engineering Research

Development of advanced algorithms and parallel software

for application of high-performance computing technology

to problems at the forefront of science and engineering

Optics and Imaging Computational Physics Computational Fluid Dynamics Ecological Modeling Chemistry and Materials Computational Biology

Quantum Optics • Optics & ImagingQuantum Optics • Optics & Imaging Image Processing and Astrophysical Observation Techniques for

Astronomy and Space Surveillance Applications (D. Tyler, S. Prasad, W. Junor, R. Plemmons, T. Schulz, J. Green, J. Seldin, P. Alsing)

Quantum Computing and Quantum Optics (I. Deutsch, C. Caves, P. Alsing, G. Brennan, J. Grondalski, S. Ghose, P. Jessen)

Optical Pulse Interactions with Nonlinear Materials (P. Bennett)

Quantum ComputingQuantum ComputingProf. Ivan Deutsch, and Prof. Carl Caves (Physics and Astronomy);

Dr. Paul Alsing (AHPCC);

Quantum Optical Lattices

By shining counter-propagating laserbeams, “crystals of light” can be formed (egg crate structures) which can be used to trap neutral atoms, e.g. cesium. By changing the phase of the light, atoms can be brought together (shift the egg crate minima) and made to interact by an additional catalysis laser. The interacting atoms form qubits and the shifting egg crate potentials act as a computer bus.

Chemistry & MaterialsChemistry & Materials

Defect Centers in a-SiO2 Using Computational Chemistry Techniques (S.P. Karna, A.C. Pineda)

Defects in Al and Cu ULSI Interconnects — Materials/Solid State Physics (S.R. Atlas, S.M. Valone, L.A. Cano)

Electron Transfer in Dendrimers (T.S. Elicker, D.G. Evans)

Dynamics at Metal Surfaces (D. Xie, H. Guo)

Molecular Dynamics of Proteins in Solution (P. Alsing, E. Coutsias)

Atom-Ion Collisions (P. Alsing, M. Riley, A. Hira)

Defects in SiODefects in SiO22

Dr. Andrew Pineda, AHPCC

Dr. Shashi Karna, AFRL

a-SiO2

p-Si

n-Si n-Si

Vdrain

Vgate

Vbias

source

a-SiO2 is the dielectric (insulator) material used in today’s semiconductor devices. Defect centers are created in manufacture and by irradiation. They are believed to be the primary charge traps in semiconductors; degrading current/voltage performance and sometimes destroying them.

• Defects are detected experimentally via EPR.Defects are detected experimentally via EPR.• Quantum mechanical (Hartree-Fock) calculations Quantum mechanical (Hartree-Fock) calculations provide detailed information candidate structure and provide detailed information candidate structure and formation mechanisms.formation mechanisms.• Same computational techniques are used to model Same computational techniques are used to model active sites of biological molecules in rational drug active sites of biological molecules in rational drug design.design.• Computations involve hundreds of electrons and Computations involve hundreds of electrons and dozens of atoms: 100’s of CPU hours on 8–32 dozens of atoms: 100’s of CPU hours on 8–32 processors of a supercomputer.processors of a supercomputer.

Molecular Dynamics simulation of the role of water in Molecular Dynamics simulation of the role of water in protein foldingprotein folding

Dr. Paul M. Alsing (AHPCC); Prof. Evangelos Coutsias (Mathematics & Statistics);

Prof. Jack McIver (Physics and Astronomy)

Visualization of large data sets frommolecular dynamics simulations in Flatland

Computational GenomicsComputational Genomics

Systems design and management Storage and manipulation of large microarray and patient datasets Database/annotation design Firewall to protect patient privacy Customized hierarchical mass storage system

Visualization Mathematical and computational analysis

Molecular classification: clustering and neighborhood analysis Identification of genetic correlations in microarray data Collaboration between biologists, medical scientists,

mathematicians, computational scientists will be essential

Computer Science ResearchComputer Science Research

Parallel Algorithms and Numerical Mathematics (D.A. Bader, P. Bennett, P. Alsing, B. Minhas)

Condor Flocking and Turing Cluster — High Throughput Computing (Z. Chen, B.T. Smith, X. Wang, M. Livny, C.D. Maestas)

Scalable Systems Lab (A.B. Maccabe)

Research Clusters: Black Bear, Vista Azul, Roadrunner (R. Ballance, P. Kovatch, J.R. Barnes, C. Maestas) — Programming Paradigms for SMP Architectures; Code Development and Optimization; Cluster Management

ActivitiesActivitiesResearch

Providing, Developing and Implementing Services Computing and visualization Distributed computing scheduling Collaborative interactive environments for researchers and training

Education and Outreach Graduate-level certificate program for students and professionals at the federal labs Native American education and training Hawaiian schools NCSA activities—educational toolkits

Training in High Performance Computing and Applications Regional industry users Federal lab users Students and faculty—local and national

Image ProcessingComputational MechanicsComputational Physics

High Performance ComputingVisualizationModeling and Simulation

Computational Chemistry

Computational Biology

R&D ProjectsR&D Projects

ProjectProject

Flatland

SMP Programming

Portals, NGIO

Access Grid Tools

CoMeT

Maui Scheduler

AreaArea

Visualization

Clusters

Networking

Collaboration

Computational Modeling

Cluster Management

Production SystemsProduction Systems

CondorCondor Distributed Workstations Remote Job Submission and Management

RoadrunnerRoadrunner Alliance Shared Computational Resource

Production Linux Cluster from Alta Technology Corporation 64 Nodes, 128 Processors, Myrinet Networking

Research SystemsResearch Systems

Black Bear Black Bear Linux Cluster Provided by VA Linux Systems

16 Nodes, 32 Processors, Myrinet Network

Vista AzulVista Azul Hybrid IBM Linux/SP with in situ Graphics

Linux: 8 Nodes, 32 Processors, Graphics-Enabled SP: 8 Nodes, 32 Processors 360 GB Storage, Shared Graphics Framebuffer