Coupling technology benchmarking in ISENES 2 S Valcke
Coupling technology benchmarking in IS-ENES 2 S. Valcke, CERFACS Coupling Workshop 2015 • • • IS-ENES 2 WP 10 -T 3 Evaluation of coupling strategies Characteristics of coupling systems: tables & mindmaps Priority coupling characteristics to benchmark Definition of the coupled benchmark suite Summary & conclusions CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 1
IS-ENES 2 WP 10 -T 3 Evaluation of coupling strategies CERFACS (9 pm), STFC (5 pm), UNIMAN (4 pm), Met. O (9 pm), DKRZ (3 pm) Objective: Define a suite of coupled benchmarks based on simplified components, which capture the essence of the coupling without the science complexity 1. Capture functional and performance characteristics of coupling system (CW 2013) 2. Code a set of simplified components reproducing these characteristics 3. Implement the coupling between these components with OASIS and ESMF. 4. Run the benchmark suite on specific platforms. 5. Analyse results and present them to the community, 6. Undertake performance modelling to support the analysis of the benchmark Ø D 10. 3 Report on benchmark suite for evaluation of coupling strategies CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 2
Characteristics of coupling systems Working groups at CW 2013, Boulder: • What are the scientific and technical requirements, including functional (e. g. , data exchange, regridding, etc. ) and non-functional (e. g. , performance, flexibility, etc. ) aspects, to build a geophysical coupled system from independent models? • What are the qualities that should be assessed in a coupling technology benchmark and how should those qualities be measured? CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 3
Characteristics of coupling systems • First series of mindmaps by US project Earth System Bridge reviewed during IS-ENES 2 Exeter workshop (Feb 2014) CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 4
Characteristics of coupling systems Coupling technologies • • • architecture: basic design principles and other general characteristics implementation: how the technology is implemented (library, parallelism, language, etc. utilities: all the utilities offered by the technology CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 5
Characteristics of coupling systems Components • characteristics of the components supported by the technology (language, parallelism, coupling data produced, …) CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 6
Characteristics of coupling systems Composition • coupling data transformation and transport, coupling modes supported CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 7
Characteristics of coupling systems Deployment how the components can be mapped on available computing resources • Metadata • as input for configuration, or produced, type of metadata CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 8
Priority coupling characteristics to benchmark ² Type of the component grid ² Number of cores used for the components - up to O(104) ² Numbers of fields exchanged ² Frequency of exchange ² Size of the coupling fields: “coarse” global grid of ~200 km resolution ² “high-resolution” global grid of 20 -50 km ² self-generated latitude-longitude grids for very high-resolution tests ² ² (Ease of use: code intrusion, development time, techniques for overcoming specific issues) CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 9
Definition of the coupled benchmark suite ² Pre-coded stand-alone components running on different grids • Fortran subroutine(s) with no physics/dynamics but real coupling characteristics • MPI parallel • Coupling fields as IN/OUT arguments, arrays in shared modules, local data • Ways to check correctness & completion • Use specific grids – latlon: latitude-longitude, arbitrary resolution (STFC/UNIMAN) – stretched: stretched & rotated, logically rectangular, e. g. ORCA (CERFACS) – icosa: quasi-uniform icosahedral, e. g. NICAM (CERFACS) – cubedsphere: quasi-uniform cubed sphere (Met. O/UNIMAN) ² ² Coupling specifications: • Coupling fields in both directions every tstep • Concurrent & sequential execution • Timings: init; 1 st tstep; mean, max, min over 99 tsteps A list of test cases to be implemented CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 10
Definition of the coupled benchmark suite ² A list of test cases to be implemented • latlon - latlon with same decomposition on both sides: – – • latlon - latlon with different decompositions on both sides: – – • • OASIS 3 -MCT : CERFACS Open. PALM : CERFACS ESMF : Met. Office OASIS 3 -MCT : Met. O ESMF : Met. O icosa - icosa : – – – • • OASIS 3 -MCT : CERFACS Open. PALM : CERFACS ESMF : CERFACS stretched –icosa – – cubedsphere (with finite differences) - latlon : – – • OASIS 3 -MCT : CERFACS Open. PALM : CERFACS ESMF : STFC/UNIMAN MCT : STFC/UNIMAN stretched - latlon: – – – • OASIS 3 -MCT : CERFACS Open. PALM : CERFACS ESMF : STFC/UNIMAN MCT : STFC/UNIMAN OASIS 3 -MCT : CERFACS Open. PALM : CERFACS stretched – Gauss reduced – OASIS 3 -MCT : CERFACS cubedsphere (with finite elements) - latlon: – – OASIS 3 -MCT : Met. O ESMF : Met. O CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 11
Summary & conclusion ² Functional and non-functional characteristics of coupling systems captured in comprehensive mindmaps ü International collaboration Ø Used as a basis to define a questionnaire to characterize existing technologies, see Rocky Dunlap’s talk ² Priority characteristics for IS-ENES 2 benchmark identified ² Coupling benchmark characteristics established: stand-alone components • coupling specifications • test-cases to be implemented • ² Work has started: Icosa & streched grid standalone components coding at CERFACS • On going- work at the Met. Offie (see Mike Hobson’s talk) • ² Benchmark tests planned 2015 -16 for deliverable 11/2016 ² Benchmarks to be extended in ESIWACE, the Centre of Excellence in Simulation of Weather and Climate in Europe, if funded CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 12
The end CW 2015, Manchester, 04/20215 - Coupling technology benchmarking in IS-ENES 2 13
- Slides: 13