slide 1 auburn university computer science and software engineering scientific computing in computer...
TRANSCRIPT
Slide 1Auburn UniversityComputer Science and Software Engineering
Scientific Computing in
Computer Science and Software Engineering
Kai H. Chang
Professor and Chair
December 5, 2014
Slide 2Auburn UniversityComputer Science and Software Engineering
Outline
High Performance Computing Architecture and I/O Optimization – Dr. Weikuan Yu Computation Software – Dr. Tony Skjellum Scientific Software Development Tools - Dr. Jeff Overbey
Slide 3Auburn UniversityComputer Science and Software Engineering
HPC Capabilities – Dr. Weikuan Yu
• Software: Unstructured Data Accelerator (UDA)Accelerator for Big Data AnalyticsTransferred to Mellanox
• In-house supercomputer: Eagles 108 nodes; InfiniBand, 10GigE and SSDGPGPU (Fermi and Kepler)Donations from Mellanox, Solarflare,
and NVIDIA
• On-campus network: TigerSphere
• Software and algorithm development2 postdoctoral researchers11 graduate students3 undergraduate studentsAlumni in prestigious labs
(IBM, ORNL, Microsoft).
Slide 4Auburn UniversityComputer Science and Software Engineering
EPIO: An Elastic Parallel I/O Frameworkfor Computational Climate Modeling
Objectives Explore and leverage computing technologies such as
parallel I/O and cloud computing for NASA’s BigData. Design new I/O approaches to managing gigantic data
from NASA climate modeling codes. Introduce elastic and non-intrusive data management
plug-in (EPIO) into the ESMF I/O framework. Integrate EPIO into representative NASA codes
(GOES-5, Model-E & GCE) for exploitation of benefits
Initial Result on the Sith Cluster (1024 nodes) at ORNL
Vertical Level 1
Vertical Level 10
Approach Evaluate and optimize data communication and I/O in
NASA climate codes (GOES-5 and Model-E) Aggregate and parallelize data for elasticity & efficiency
GEOS-5
History
GrDAS NetCDF4/HDF5 EPIO
Global Climate Model
AGCM OGCM
Atmosphere Analysis
New I/O component
Impact Provide new processing and analytics approaches
for climate data management.
Enable streamlined data movement through workflows of climate simulation, analytics and visualization systems.
Enable portable and efficient NASA climate codes on petascale National Leadership Computing Facilities, and prepare them for future exascale.
Slide 5Auburn UniversityComputer Science and Software Engineering
Areas for Potential R&D Collaboration - Dr. Tony Skjellum
Over 30 years of experience formulating and parallelizing programs Long-term R&D in scalable parallel mathematical libraries Strong experience with simulation codes (e.g., based on PETSc,
integration engines, and homegrown) Leading R&D in MPI standardization, implementation, and utilization Additional R&D and research skills in CUDA, OpenCL, OpenMP,
SSE/AVX, and cluster computing Strengths in algorithmic optimization and systems programming On-going research in fault tolerant algorithms and parallel
programming Mixed-language programming expertise (C, C++, Fortran) Experience with cache-friendly and data distribution independent
algorithms
Slide 6Auburn UniversityComputer Science and Software Engineering
Software Development Tools for Scientific Computing- Dr. Jeff Overbey
• Improving the Eclipse integrated development environment to support scientific software development
– Project lead: Photran (FortranDevelopment Tools for Eclipse)
– Committer: PTP (EclipseParallel Tools Platform)
– June release: 184,846 downloads
– Commercially: IBM ParallelEnvironment Developer Edition
– Collaboration withIBM, UNLP, NCSA, LSU, Oregon
• Refactorings for Fortranlegacy code migration
• Refactorings for GPU computing
Slide 7Auburn UniversityComputer Science and Software Engineering
Other Related Expertise Modeling Simulation - Dr. Levent Yilmaz Operating Systems – Dr. Xiao Qin Databases – Dr. Jeff Ku
Slide 8Auburn UniversityComputer Science and Software Engineering
Questions?