Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
Current Status of High Performance Computing and
CMIP6 at CMA
WANG Bin, WU Huanping, Xin XiaogeNMIC/CMA, NCC/CMA
Sept.12, 2019. ICAS 2019, Stresa, Italy.
Outline
• High Performance Computing at CMA
• Current Status of CMIP6 at CMA
NMIC in CMA——One of CMA’s National Operational Centers
Responsible for :
• Meteorological Communication
• Meteorological Data
• Meteorological Supercomputing
• Meteorological Administrative Affairs
Missions:
• Meteorological Communication Center: to collect and disseminate globally exchanged data, domestic observations and forecast products.
• Meteorological Data Center: to conduct life cycle data management (QC, Processing, Storage, Services and Archiving); develop atmospheric research datasets.
• High Performance Computing Center: to provide HPC resources and application support for operations, researches & development of numerical prediction models.
CMA Offices Met. Service
GovernmentDisaster Mitigation
Social Sectors Economy
CitizensBetter livelihood
High-Quality Data, Value-added ServiceFor Disaster Mitigation, Social Economy & Citizen Livelihood
VISION:
CMA Meteorological Data Center
CommunicationCenter
HPCCenter
Global Met. Data Domestic Met. Obs
Gov. Departments Crowd-sourcing Data
NWP Models Climate Models
Atmos. Reanalysis
Common Interface
Specialized Models
History of CMA HPC
PI-Sugon
• Peak Capability : 8,198.5 TFLOPS• Storage Capacity : 23,088 TB• CPU Cores : 98,432• PUE : 1.23
Two Subsystems backup each other:• Computing Nodes : 1504 • CPU + GPU• Intel MIC
• 2 Intel Xeon Processors per Node• 100Gb/s InfiniBand EDR network• Parastor 300 Parallel File System
Software Stack
IBM HPC PI-Sugon
Improvement
Peak Performance ~1000 TFLOPS ~8000 TFLOPS
Storage Capacity ~4000 TB ~18000 TB
Inter -Connection QDR 40Gb/s EDR 100Gb/s
Difference
OS AIX 7.1.0.0 RedHat Enterprise 7.4
并行文件系统 IBM GPFS Sugon Parastor
作业调度软件 IBM LSF(Loadleveler) Sugon Gridview(Slurm)
集群管理 IBM xCAT Sugon Clusconf
应用软件巨大差异!!!
编译器IBM XL C/C++ Compiler V11;
IBM XL Fortran Compiler V13.1
GNU(C/C++/Fortran等)
Intel Parallel Studio XE Cluster Edition
PGI编译器
并行环境 IBM Parallel Environment Runtime EditionOpenMP
OpenMPI,MPICH2,Mvapich2等
IBM & PI-Sugon
PI-Sugon Resource Usage
GRAPES & BCC_CSM
• GRAPES = Global/Regional Assimilation PrEdictionSystem
• BCC_CSM = Beijing Climate Center Climate System Model
GRAPES-GFS GRAPES_MESO GRAPES-TYM GRAPES-GEPS GRAPES-MEPS
Forecast range 10d 1.5d 5d 15d 3.5d
Domain Global China West Pacific Global China
H-resolution 25KM 3KM 9KM 50KM 10KM
V-resolution60L
3hPa
50L
10hPa
68L
10hPa60L 3hpa
50L
10hPa
Forecast time00,12 UTC 240h
06,18 UTC 120h00,06,12,18 UTC 00,12 UTC
00,12UTC
31members
00,12 UTC
15 members
Operation Runtime Schedule
12
• High performance computer management software
• Refined resource management system
• Operational monitoring system
• Numerical Model R & D Supporting Software
• Code management system
• GRAPES Integrated Setting Experiment Tool(GISET)
Model-Supportive Software Systems
Refined resource management system
• Resource management of IBM & PI-Sugon systems
• Unified management of national and regional resources
• Real-time and historical statistical analysis of system resource usage and utilization
• Computing resource and storage resource usage accounting
• Model & job statistical analysis
• Monitoring of IBM & PI-Sugon systems and software,audio alarm
• Unified management of national and regional resources
• Real-time monitoring and historical statistical analysis of failure
• Automatic reporting of system availability & statistical analysis of failure
• Fault handling workflow & fault knowledge database
• Model job monitoring
• Real-time monitoring of memory, CPU utilization and jobs
Operation monitoring system
• Code Management System -Perforce(since 2010)
• P4 using on IBM and PI-Sugon HPC
• GRAPES_GFS,GRAPES_MESO,BCC_CSM code repository
• National & regional distributed design for GRAPES_MESO collaboration
• Code version control and integration control
• Git-based Code Management System-METCODE(since 2019)
• Github-liked web remote repository and local repository,easier to share.
METCODE
• GRAPES Integrated Setting Experiment Tool
• Experiment construction
• Experiment scheduling (ecFlow)
• Experiment sharing, statistics, compare
• Integrated code and experiment data
management
• Design and implementation based
on C/S mode
• Coded by python
• Back-end services run on servers
GRAPES Integrated Setting Experiment Tool(GISET)
Programming Model on TaihuLight
Development on Sunway TaihuLight
Efforts are made to exploit many-core acceleration technology
Outline
• High Performance Computing at CMA
• Current Status of CMIP6 at CMA
19
State Council
Headquarters
Office
Department ofIntegrated
Observations.
Department of EmergencyResponse,Disaster MitigationandPublic Services
Department of Science & Technology and Climate Change
Department of Human Resources
Department of Planning&Finance
Department of Policy &Regulations
Department of InternationalCooperation
National Met.Centre/NWP Centre
National
Sat. Met. Centre
Chinese
AcademyofMet. Sciences
CMA
Training
Centre
Met.press
Provincial Met. Services 31
China Met.Society
Specialized Research Institutions
Under CMA (8)
Prefecture Met. Offices 329
County Weather Offices 2155
ChinaMet.Newspaper
CMA
National ClimateCentre
NationalMet.InfoCentre
CMAMet. Observation Centre
Department of Forecasting and Networking
WMO Regional Climate Centre
CMA Climate Change Centre
Public Service Centre
CMA Communication and Outreach Centre
CMA Asset Operation Centre
Regional Climate Centres
BCC in CMA
BCC Organization
National Climate Center (NCC)
Beijing Climate Center (BCC)
General Office (GO)
6 FTE
Division of Operation, Science and Technology
(DOST)
7 FTE
Division of Personnel Affairs (DPA)
4 FTE
Climate System Modeling Division (CSMD)
27FTE
Climate Services Division (CSD)
16 FTE
Laboratory for Climate Studies (LCS)
22 FTE
Climate Prediction Division (CPD)
31 FTE
Meteor. Disaster Risk Management Division
(MDRMD)
20 FTE
Climate Change Division (CCD)
24 FTE
Operational System Management Division (OSMD)
15FTE
FTE:Full Time Employee
Total :196Climate Application
8 FTE
Smart clouds(Hardware)
Climate data & products center
CIPAS(climate monitoring and prediction system)
Climate disaster risk management system
Climate application service platform(CLAP)
APP Web(NCC)Blog/Wechat
Op
era
tion
Sta
nd
ard
s
Tech
no
log
y fra
mew
ork
Climate modelsystems
Climate operational simulation platform
Overview of Climate Operational Systems
BCC-ESM1.0 BCC-CSM2-MR BCC-CSM2-HR
BCC_AGCM3(T42L26,Top:2.19 hPa)Top: 2.19 hPaBCC_AVIM2.0(T42)MOM4-L40v2(1/3~1°)SIS(1/3~1°)BCC-AGCM3-Chem
BCC_AGCM3(T106L46, Top:1.46hPa)BCC_AVIM2.0(T106)MOM4-L40v2(1/3~1°)SIS(1/3~1°)
BCC_AGCM3(T266L56: Top:0.1hPa)BCC_AVIM1.0(T266)MOM4-L40v2 (1/3~1°)SIS(1/3~1°)
middle resolution (110km)low resolution (280km) high resolution (45km)
Climate models developed by BCC
CMIP6: 21
About 6 staffs get involved in NCC
Computing and storage resource
roughly 50,000,000 core hours and200TB
CMIP6 Mips being carried out with BCC-CSM2-MR
Short name of MIP Model version Status
1 DECK BCC-ESM1 and BCC-CSM2-MRFinished
historicalBCC-ESM1 and BCC-CSM2-MR Finished
2 ScenarioMIP BCC-CSM2-MR Finished Tier1
3 C4MIP BCC-CSM2-MR Finished Tier1
4 CFMIP BCC-CSM2-MR Finished Tier1
5 DAMIP BCC-CSM2-MR Finished Tier1
6 GMMIP BCC-CSM2-MR Finished Tier1
7 LS3MIPBCC-AVIM2,
BCC-CSM2-MRFinished Tier1
8 LUMIP BCC-CSM2-MR Finished Tier1
9 DCPP BCC-CSM2-MR Ongoing
10AerChemMIP BCC-ESM1
Finished Tier1
11 HighResMIP BCC-CSM2-HR Ongoing
Tier 1 experiments of MIPS will be carried out.
BCC_AGCM2 ---> CMIP5 BCC_AGCM3 --- > CMIP6
Originated from CAM3
A component of BCC-CSM1.1, BCC-CSM1.1m
Resolution: T42L26, T106L26
Model Dynamic Core:
A modified dynamic framework for atmospheric spectral model (Wu et al., 2008: J.Atmos.Sci.)
Model Physics
Deep convection:Wu T., 2012: A Mass-Flux Cumulus Parameterization Scheme for Large-scale Models: Description and Test with Observations, Clim. Dyn., 38
Dry Adiabatic adjust scheme
Snow cover fraction parameterization (Wu T. and Wu G., 2004)
A modified sensible and latent flux parameterization on the ocean- Atmosphere interface (Wu et al. 2010: Climate Dynamics)
No indirect effects of aerosols
Ref: Wu et al. 2010: Climate Dynamics
A component of BCC-CSM2-MR, BCC-CSM2-HR
Resolution: T106L46, T266L56
Model Dynamic Core:
The spatially variant divergence damping scheme in higher resolution version (Whitehead et al., 2011)
Model Physics:
A gravity wave drag generated by orography, and convection (Beres et al., 2004)
A modified Wu’2012 deep convective scheme
A new scheme to parameterize deep and shallow cumulus cloud amount.
A modified parameterization scheme for surface turbulent fluxes between air and ocean/sea ice
Indirect effects of aerosols. The liquid cloud droplet number concentration is diagnosed using the aerosols mass
AGCM component
Time series of global (60°S to 60°N) mean surface air temperature from 1850 to 2015. The reference climate is for the period of 1961 to 1990.
Historical simulation of global SAT and precipitation
Summer mean precipitation climatology during 1980-2005.
BCC-CSM2-MR
BCC-CSM1.1m
GPCP
Summer precipitation in China
BCC-CSM1.1m 320x160(1.125x1.12)CCSM4 288x192 (1.25x0.95)CESM1(BGC) 288x192 (1.25x0.95)CESM1(CAM5) 288x192 (1.25x0.95)CMCC-CM 480x240(0.75x0.75)CNRM-CM5 256x128 (1.4x 1.4)MIROC4h 640x320(0.5625x0.56)MIROC5 256x128(1.4x1.4)MRI-ESM1 320x160(1.125x1.12)MRI-CGCM3 320x160(1.125x1.12)
The rainfall amount in southeast China is larger than BCC-CSM1.1m, closer to OBS than BCC-CSM1.1m.
BCC-CSM-MR can well simulate the distributions of East China summer precipitation. The spatial correlation is among the top four models.
BCC-CSM2-MR BCC-CSM1.1m
CMIP5 MME OBS
CMIP5 models with resolution higher than 1.5°
East Asian Summer Monsoon
850hPa winds (vectors) and U200 (shaded).
East Asian Monsoon Skill
BCC-CSM2-MR
BCC-CSM1.1m
CMIP5 MME9
JRA55
BCC-CSM-MR improves the skill in the simulation of East Asian summer monsoon relative to BCC-CSM1.1m.
The skill of BCC-CSM2-MR is among the top four models.
Projection of temperature and precipitation change in 21 century over China and US
with BCC-CSM2-MR in SSPs
Relative to the 20-yr mean of piControl
Climate sensitivity of BCC-CSM2-MR
Projection of global mean temperature and precipitation
Global mean SAT
Global mean land precipitation
Warming in long-term (2081-2100)
SSP585: 3.3℃SSP370: 3.0℃SSP245: 1.8℃SSP126: 0.9℃
Rainfall change in long-term(2081-2100)
SSP585: 0.10 mm/daySSP370: 0.09 mm/daySSP245: 0.06 mm/daySSP126: 0.03 mm/day
One of the early online Models
https://pcmdi.llnl.gov/CMIP6/ArchiveStatistics/esgf_data_holdings/
CMIP6 status: data availability
Check status at PCMDI website below
Online product of BCC models