Jack Dongarra

1

• Jack Dongarra
• University of Tennessee
• http//www.cs.utk.edu/dongarra/

2
ICL Four Thrust Areas

• Heterogeneous Distributed Computing
• Numerical Linear Algebra
• Software Repositories
• Performance Evaluation
• Helping to Formulate Standards
• Software and ideas have found there way into many
  areas of Computational Science
• Between 40-50 people At the moment...
• 20 Researchers Research Assoc/Post-Doc/Research
  Prof
• 20 Students Graduate and Undergraduate
• 6 Support staff Secretary and Systems

3
Computer Science Department

• Numerical Linear Algebra Algorithms and Software
• EISPACK, LINPACK, BLAS, LAPACK, ScaLAPACK, PBLAS,
  Templates, ATLAS
• Heterogeneous Network Computing
• PVM, MPI, NetSolve, I2-DSI, Harness
• Software Repositories
• Netlib, NA-Digest, NHSE, RIB
• Performance Evaluation
• Linpack Benchmark, Top500, PAPI

Employment and academic opportunities in the ICL
http//icl.cs.utk.edu
4
Numerical Linear Algebra

• LAPACK, ScaLAPACK
• Software libraries for performing dense and band
  linear algebra computations on shared memory and
  distributed-memory computers
• Portable, efficient, standard
• ATLAS
• Automatic generation and optimization of
  numerical software for processors with deep
  memory hierarchies
• Templates
• State of the art algorithms and wisdom on
  iterative methods for large sparse linear systems
  and eigenvalue problems

5
Heterogeneous Network Computing

• NetSolve
• Network based computational server that allows
  users to access computational resources
  distributed across the network.
• I2-DSI
• Designing and deploying a reliable, scalable
  high-performance storage services infrastructure
• Harness
• Heterogeneous Adaptable Reconfigurable Networked
  Systems
• Plugability
• connectibility

6
Software Repositories

• Netlib
• Central repository for mathematical software,
  papers, and databases
• NHSE
• An interface to a distributed collection of HPCC
  repositories.
• Repository in a Box (RIB)
• A toolkit for creating software repositories that
  can interoperate with other software repositories
  via the Internet.

7
Performance Evaluation

• Linpack benchmark
• Performance benchmark
• Top500
• Collection of the sites that have the 500 fastest
  supercomputers worldwide.
• PAPI
• Performance API for monitoring applications
  programs

8
Funding

• NSF
• CISE
• Applied Math
• DOE
• MICS
• ASCI
• DOD
• Modernization Program
• NASA
• IPG from Ames/NAS

• Microsoft Research
• IBM
• Sun
• Intel
• HP
• State of Tennessee

9
Collaboration

• ORNL
• UCB/UC Davis
• UCSD
• Globus/ANL/ISI
• Salk Institute
• Danish Technical University/UNIC
• ETHZ, Zurich
• ETL Tsukuba, Japan
• Thailand, Kasetsart U

10
Resources

• 26 Cluster Machines in Torc
• 3 Alphas, 1 Apple Mac, and 22 Pentium
• Network Connections include 100Mbit, Myrinet,
  Gigabit, Giganet
• OS's include Linux, Windows, Compaq True64,
  MacLinux, and Mac OS
• 48 Desktop machines
• 20 Server machines (Compute, Services, Storage)
• 34 Laptops
• Of these, 30 machines are donations/loaners

11
Futures

• NSF ST Center
• Join with Rice, UCSD, ANL, ISI, IU, UIUC
• UTK Center of Excellence
• 5M Center focus on information technology
• NSF ITR effort
• A number of proposals

12
Schedule

• Monday, May 22
• 900 Meet with Shirley, Scott, and Terry
• 1000 Jack -- Overview of ICL
• 1100 Year 5 projects - Shirley, Scott, Jeremy
• 1200 Lunch
• 115 Richard Luczak
• 145 PAPI - Nathan, Kevin
• 230 NetSolve - Michelle, Susan
• 315 Metacomputing evaluation - Graham, David
• 400 ATLAS and HPC/Linear Algebra - Clint,
  Antoine
• Tuesday, May 23
• 900 Meet with Shirley, Scott, and Tracy
• 1030 MPI-Connect, Parallel/remote I/O - Graham
• 1115 Benchmarking and Performance Modeling -
  Erich
• 1200 Lunch