research computing initiative
DESCRIPTION
Research Computing Initiative. Dimitri Mavriplis Department of Mechanical Engineering Alex Buerkle Department of Botany Dan Stanescu Department of Mathematics. Research Computing at UW. The RCI fills various roles EPSCoR RII calls for comprehensive cyberinfrastructure plan - PowerPoint PPT PresentationTRANSCRIPT
Research Computing Initiative
Dimitri MavriplisDepartment of Mechanical Engineering
Alex BuerkleDepartment of Botany
Dan StanescuDepartment of Mathematics
Research Computing at UW
• The RCI fills various roles– EPSCoR RII calls for comprehensive
cyberinfrastructure plan• RCI fills this role
– Computational Science is a strategic area of emphasis at UW
• RCI will provide focus across campus• RCI will leverage current and growing UW
strengths in this area
Research Computing at UW
• The RCI is research centric– Cyberinfrastructure to further research goals
of ESICT (FOSTER/SOLPIN)• The RCI has an academic component
– Interdisciplinary minor in Computational Science
RCI for ESICT
• Provide hardware/software and expertise to support SOLPIN, FOSTER and other experiments – Networks : Data acquisition, retrieval,
dissemination– Mass storage facility: Data archiving, mining– High performance computing hardware:
Simulation as complement to observational data
RCI across Campus
• UW has targeted computational science as area of emphasis– Growing importance in all science areas– Energy related areas in particular– Various recent faculty hires– NCAR Petaflops center
• Campus wide facility/expertise required to advance to further this emphasis
Academic Component : ICS Minor
• Undergraduate minor in computational science approved and underway, Graduate minor degree program under formation– Core required courses combined with elective
courses– Requires development of various new courses– Interdisciplinary nature is key
• Mathematics, computer science, science domain
• Training future computational scientists is recognized as a national imperative
RCI Hardware
• Fast networks and good connectivity for SOLPIN/FOSTER
• Petabyte storage facility• High performance computing (HPC)
platform – e.g. 2000 cpus– Necessary to bridge gap between ad-hoc
departmental clusters and national supercomputing facilities (NCAR, others)
RCI Personnel
• 3 support staff positions– Head System administrator/HPC specialist
• Administer RCI facility, provide HPC expertise
– Assistant System administrator/HPC specialist• Support facility, scientific libraries, training
– HPC Application support position• Application domain HPC specialist
• Positions report to IT Department• Overseen by RCI Director (Faculty member)
RCI Related Faculty Positions
• Request 2 faculty positions through EPSCoR RII– Support for startup packages
• Faculty Areas of Specialization:– 1: Computer Architectures for High Performance
Computing • (Dept. of Computer Science )
– 2: Computational Science• Broad search, Earth science related• Joint appointment possible
– e.g. science domain, math or computer science
Oil, Water & RockProperties
(Modified from Gordon Adamson et al, Oilfield Review, 1996)
Predictive model
EOR Injected FluidProperties
History Matching
Fine-scale static model
Up-scaled static model
Production DataWell Configurations
Reservoir Simulation (EORI)
Fully dynamic particle-based modeling of three-phase flow in a voxel image of the rock (Dept. Chem/Petroleum Eng)
(Figure by Bakke, 1997 and Øren 1998, 2002)
• Smoothed Particle Hydrodynamics
• Lattice Boltzmann Methods
• Experimental Scanning Facilities currently being acquired
Computational biology at UW(c/o A. Buerkle, Botany)
• Reconstruction of evolutionary relationships and ancestral states among organisms (phylogenetics).
• Comparative and population genomics to discover genes of significance to human health, agriculture and adaptation.
• Modeling of 3D protein structure.
Research areas include:
Computational tasks:
Partial representation of mammalian supertree (from Bininda-Emonds et al. 2007, Nature 446:507)
• Estimation of evolutionary trees of protein structures are large optimization problems
• Many inferences are based on large sets of (often parallel) simulations
• UW work currently on mid-size clusters. Ripe for migration to higher capacity hardware
Computational Aeroacoustics Dan Stanescu (Math)
• Aeroacoustic Simulations using High-Order Spectral Difference Methods• 22M pts, 6th order accurate in space• Run on up to 512 CPUS (IBM Power4)
Round Jet Simulation(Dan Stanescu)
• Iso-vorticity surfaces for round jet flow• Mesh size 40 x 30 x 30 , p = 4 • Reynolds number = 100,000, Mach = 0.15
Scalability of Unstructured Mesh Multigrid Solver on NASA Columbia Supercomputer
• D. Mavriplis, M. Aftosmis, M. Berger: High-Resolution Aerospace Applications on the NASA Columbia Supercomputer, Best paper award, SC05, Seattle, Nov 2005
• Blue Gene/L, Cray XT-3 Benchmarks.
72 million grid points
Unstructured Mesh Finite Volume Multigrid
Requires combined MPI/OMP: 5.2 Tflops
Summary
• Research Computing Initiative will provide critical support for the ESICT project– Indispensible cyberinfrastructure component
• RCI will result in an integrative strategy for research computing across campus– Focus and support for research programs– Coordinate educational efforts