NERSC SSP

From View

Jump to: navigation, search

Contents

Purpose of Benchmark

The NERSC Sustained System Performance (SSP) metric ( http://www.nersc.gov/projects/SDSA/software/?benchmark=ssp ) is used to measure the execution performance and compiler capabilities of existing and future supercomputer systems and to provide an ongoing glimpse of a portion of the NERSC scientific computing workload. Each of the benchmark programs represents a particular subset and/or characteristic of the NERSC workload, which generally consists of solving complex scientific problems using a variety of state-of-the-art computational techniques. There is keen interest on the part of the NERSC SDSA ( http://www.nersc.gov/projects/SDSA/software/ ) team to have vendor and university researchers use these codes to help guide future system designs.

The SSP benchmarks evolve over time to reflect changes in the NERSC workload and evolution in the state-of-the-art scientific codes that are run on HPC systems.

Target Platforms(s)

The SSP is updated for each new procurement to target each new generation of HPC system architectures. It evolves to accomodate both new system architectures and trends in HPC application design.

Application Domain(s)

The SSP is targeted at scalable parallel scientific applications that run on the largest available HPC resources such as NERSC ( http://www.nersc.gov/ )supercomputing center and any of the centers that are represented on the Top500 ( http://www.top500.org ) list of largest HPC systems in the world.

The SSP attempts to cover a broad array of application domains and classes of algorithms. The current benchmark suite includes coverage of the following application areas and computational methods.

   CAM: CCSM Community Climate Model (Finite Volume method on regular grid)
   GAMESS: Computational Chemistry (particle method)
   GTC: 3D Gyrokinetic Toroidal Code (Particle in Cell method on irregular mesh)
   MADBench: Microwave Anisotropy Dataset Computational Analysis Benchmark (dense linear algebra + out-of-core I/O)
   MILC: MIMD Lattice Computation (4D lattice)
   PARATEC: Parallel Total Energy Code (density functional theory / spectral method)
   PMEMD: Particle Mesh Ewald Molecular Dynamics (particle mesh ewald)

Current status

Number of Benchmarks

The current NERSC-5 SSP consists of 7 application codes ( http://www.nersc.gov/projects/SDSA/software/?benchmark=NERSC5 ).

Scale of Adoption

The SSP is primarily used by NERSC to assess competing HPC systems for procurements, but elements of the SSP have been adopted by other agencies such as the DoD MSRCs for the TI-06 benchmark suite and for the NSF Track-1 and Track-2 HPC system competitions.

Personal tools