Current Efforts of SPEC HPG

1

• Current Efforts of SPEC HPG
• Application Benchmarks for High Performance
  Computing
• IPSJ SIGMPS 2003
• Matthias MuellerHigh Performance Computing
  Center Stuttgart mueller_at_hlrs.de
• Kumaran Kalyanasundaram, G. Gaertner, W.
  Jones, R. Eigenmann, R. Lieberman, M. van
  Waveren, and B. Whitney
• SPEC High Performance Group

2
Outline

• What is SPEC and SPEC HPG?
• Why do we need benchmarks?
• Benchmarks currently produced by SPEC HPG
• What do we need for the future?

3
What is SPEC?

• The Standard Performance Evaluation Corporation
  (SPEC) is a non-profit corporation formed to
  establish, maintain and endorse a standardized
  set of relevant benchmarks that can be applied to
  the newest generation of high-performance
  computers. SPEC develops suites of benchmarks and
  also reviews and publishes submitted results from
  our member organizations and other benchmark
  licensees.
• For more details see http//www.spec.org

4
SPEC Members

• Members3DLabs Advanced Micro Devices Apple
  Computer, Inc. ATI Research Azul Systems,
  Inc. BEA Systems Borland Bull S.A. Dell
  Electronic Data Systems EMC Encorus
  Technologies Fujitsu Limited Fujitsu Siemens
  Fujitsu Technology Solutions Hewlett-Packard
  Hitachi Data Systems IBM Intel ION
  Computer Systems Johnson Johnson Microsoft
  Mirapoint Motorola NEC - Japan Network
  Appliance Novell, Inc. Nvidia Openwave
  Systems Oracle Pramati Technologies PROCOM
  Technology SAP AG SGI Spinnaker Networks
  Sun Microsystems Sybase Unisys Veritas
  Software Zeus Technology

5
SPEC HPG SPEC High-Performance Group

• Founded in 1994
• Mission To establish, maintain, and endorse a
  suite of benchmarks that are representative of
  real-world high-performance computing
  applications.
• SPEC/HPG includes members from both industry and
  academia.
• Benchmark products
• SPEC OMP (OMPM2001, OMPL2001)
• SPEC HPC2002 released at SC 2002

6
Currently active SPEC HPG Members

• Fujitsu
• HP
• IBM
• Intel
• SGI
• SUN
• UNISYS
• University of Purdue
• University of Stuttgart

7
Where is SPEC Relative to Other Benchmarks ?

• There are many metrics, each one has its purpose
• Raw machine performance Tflops
• Microbenchmarks Stream
• Algorithmic benchmarks Linpack
• Compact Apps/Kernels NAS benchmarks
• Application Suites SPEC
• User-specific applications Custom
  benchmarks

Computer Hardware
Applications
8
Why do we need benchmarks?

• Identify problems measure machine properties
• Time evolution verify that we make progress
• CoverageHelp the vendors to have representative
  codes
• Increase competition by transparency
• Drive future development (see SPEC CPU2000)
• Relevance Help the customers to choose the
  right computer

9
Comparison of different benchmark classes
10
SPEC OMP

• Benchmark suite developed by SPEC HPG
• Benchmark suite for performance testing of shared
  memory processor systems
• Uses OpenMP versions of SPEC CPU2000 benchmarks
• SPEC OMP mixes integer and FP in one suite
• OMPM is focused on 4-way to 16-way systems
• OMPL is targeting 32-way and larger systems

11
SPEC OMP Applications
Code Applications Language
lines ammp Molecular Dynamics C
13500 applu CFD, partial LU
Fortran 4000 apsi Air pollution
Fortran 7500 art Image Recognition\
neural networks
C 1300 fma3d Crash simulation
Fortran 60000 gafort Genetic
algorithm Fortran 1500 galgel CFD,
Galerkin FE Fortran 15300 equake
Earthquake modeling C 1500 mgrid
Multigrid solver Fortran 500 swim
Shallow water modeling Fortran 400
wupwise Quantum chromodynamics Fortran
2200  
12
CPU2000 vs OMPL2001
13
SPEC OMP Results

• 66 submitted results for OMPM
• 24 submitted results for OMPL

14
SPEC OMPL Results Applications with scaling to
128
15
SPEC OMPL Results Superlinear scaling of applu
16
SPEC OMPL Results Applications with scaling to 64
17
SPEC HPC2002 Benchmark

• Full Application benchmarks(including I/O)
  targeted at HPC platforms
• Currently three applications
• SPECenv weather forecast
• SPECseis seismic processing, used in the search
  for oil and gas
• SPECchem comp. chemistry, used in chemical and
  pharmaceutical industries (gamess)
• Serial and parallel (OpenMP and/or MPI)
• All codes include several data sizes

18
SPEC ENV 2002

• Based on the WRF weather model, a
  state-of-the-art, non-hydrostatic mesoscale
  weather model, see http//www.wrf-model.org
• The WRF (Weather Research and Forecasting)
  Modeling System development project is a
  multi-year project being undertaken by several
  agencies.
• Members of the WRF Scientific Board include
  representatives from EPA, FAA, NASA, NCAR, NOAA,
  NRL, USAF and several universities.
• 25.000 lines of C and 145.000 lines of F90

19
SPEC ENV2002

• Medium data set SPECenvM2002
• 260x164x35 grid over Continental United States
• 22km resolution
• Full physics
• I/O associated with startup and final result.
• Simulates weather for a 24 hour period starting
  from Saturday, November 3nd, 2001 at 1200 A.M.
• SPECenvS2002 provided for benchmark researchers
  interested in smaller problems.
• Test and Train data sets for porting and
  feedback.
• The benchmark runs use restart files that are
  created after the model has run for several
  simulated hours. This ensures that cumulus and
  microphysics schemes are fully developed during
  the benchmark runs.

20
SPECenv execution models on a Sun Fire 6800
Medium scales better OpenMP best for small
size MPI best for medium size
21
SPEC HPC2002 Results SPECenv scaling
22
SPECseis execution models on a Sun Fire 6800
Medium scales better OpenMP scales better than MPI
23
SPEC HPC2002 Results SPECseis scaling
24
SPECchem execution models on a Sun Fire 6800
Medium shows better scalability MPI is better
than OpenMP
25
SPEC HPC2002 Results SPECchem scaling
26
Hybrid Execution for SPECchem
27
Current and Future Work of SPEC HPG

• SPEC HPC
• Update of SPECchem
• Improving portability, including tools
• Larger datasets
• New release of SPEC OMP
• Inclusion of alternative sources
• Merge OMPM and OMPL on one CD

28
Adoption of new benchmark codes

• Remember that we need to drive the future
  development!
• Updates and new codes are important to stay
  relevant
• Possible candidates
• Should represent a type of computation that is
  regularly performed on HPC systems
• We currently examine CPU2004 for candidates
• Applications from Japan are very welcome
  !!!Please contact SPEC HPG or me
  ltmueller_at_hlrs.degt if you have a code for us.

29
Conclusion and Summary

• Results of OMPL and HPC2002
• Scalability of many programs to 128 CPUs
• Larger data sets show better scalability
• Best choice of programming model (MPI,OpenMP,
  hybrid) depends on
• Hardware
• Program
• Data set size
• SPEC HPC will continue to update and improve the
  benchmark suites in order to be representative of
  the work you do with your applications!

30

• BACKUP

31
SPEC Members

• Members3DLabs Advanced Micro Devices Apple
  Computer, Inc. ATI Research Azul Systems,
  Inc. BEA Systems Borland Bull S.A. Dell
  Electronic Data Systems EMC Encorus
  Technologies Fujitsu Limited Fujitsu Siemens
  Fujitsu Technology Solutions Hewlett-Packard
  Hitachi Data Systems IBM Intel ION
  Computer Systems Johnson Johnson Microsoft
  Mirapoint Motorola NEC - Japan Network
  Appliance Novell, Inc. Nvidia Openwave
  Systems Oracle Pramati Technologies PROCOM
  Technology SAP AG SGI Spinnaker Networks
  Sun Microsystems Sybase Unisys Veritas
  Software Zeus Technology
• AssociatesArgonne National Laboratory CSC -
  Scientific Computing Ltd. Cornell University
  CSIRO Defense Logistics Agency Drexel
  University Duke University Fachhochschule
  Gelsenkirchen, University of Applied Sciences
  Harvard University JAIST Leibniz
  Rechenzentrum - Germany Los Alamos National
  Laboratory Massey University, Albany NASA
  Glenn Research Center National University of
  Singapore North Carolina State University PC
  Cluster Consortium Purdue University Queen's
  University Seoul National University Stanford
  University Technical University of Darmstadt
  Tsinghua University University of Aizu - Japan
  University of California - Berkeley
  University of Edinburgh University of Georgia
  University of Kentucky University of Illinois -
  NCSA University of Maryland University of
  Miami University of Modena University of
  Nebraska - Lincoln University of New Mexico
  University of Pavia University of Pisa
  University of South Carolina University of
  Stuttgart University of Tsukuba Villanova
  University Yale University

32
CPU2000 vs. OMPM2001
33
CPU2000 vs OMPL2001
34
Program Memory Footprints
35
SPEC ENV2002 data generation

• The WRF datasets used in SPEC ENV2002 are created
  using the WRF Standard Initialization (SI)
  software and standard sets of data used in
  numerical weather prediction.
• The benchmark runs use restart files that are
  created after the model has run for several
  simulated hours. This ensures that cumulus and
  microphysics schemes are fully developed during
  the benchmark runs.