View
217
Download
0
Tags:
Embed Size (px)
Citation preview
The Realization of GRAPES Model on
China Meteorological Application Grid
Xuesheng YANGChinese Academy of Meteorological Sciences
China Meteorological [email protected]
2005/09/16
Characteristics of Numerical weather prediction
• Meteorology is one of the main components in the grid application– US:
• LEAD (Linked Environment for Atmospheric Discovery)• MEAD (Modeling Environment for Atmospheric Discovery)
– ECMWF: ECAccess
• NWP system includes preprocessing, analysis, numerical model, post processing, visualization:– execution processes are dynamically connected – enormously complex even if run individually – Involves in data management, remote collaborative research, hea
vy computation, massive data and intensive services• to ensure the proper operation of a NWP:
– massive computing resources needed
Existed issues
• tough issues meteorologists encountered:– organization of the NWP workflow and data– execution of model simulations– management of the resulting large volumes of data– subsequent data analysis– visualization
• model developers and computing resources stationed in different locations– telephone, mobile, e-mail, …
• inadequate for conducting a research project that involves in huge amount of simulations and extensive on-line discussions and consultations
– Computing resources can’t satisfy the research requirements• divert researchers’ focus on how to acquire computing resources
instead of addressing the scientific issues
Existed computing resources at CMA
• Available operational computing resources:– IBM Cluster 1600 – IBM SP
• Available computing resources for research:– IBM – Clusters: Sunway, Dawning, Oscar, Ocean
• CAMS• NMIC• Guangdong• Shanghai• Guangxi • etc.
• do not get full utilized– Computing resources– manpower
2. China Meteorological Application Grid(CMAG)
• grid portal
• GridWeather– a NWP workflow control interface
• GRAPES meso-scale model
• user management sub-system
• CVSExplorer – a code management system
• visualization
Tomcat5.28 Application Server
GridSphere Container
Base Portlets
LoginP
ortlet
ProfileM
anager
PortletA
pplication
LayoutM
anager
GridUser PortletsPortlets Modules
CredentialR
equest
CredentialS
ign
CredentialD
eleteG
ridUserM
ap
CMAGPortal Page
GridWeather portletsPortlets Modules
CredentialR
ecieve
JobSubm
ission
FileM
anager
FileT
ransfer
ResourceB
rowser
JobMonitoring
JobPosttreatm
ent
Forecast PortletsPortlets Modules
RainF
orecast
CityF
orecast
Elem
entForecast
PrecipitionO
bserved
CVSExplorerPortlets Modules
ProjectM
anager
CodeE
xplorer
CV
SM
anager
GridServices Myproxy GRAPES service
Grid Software(GT2,GT3,GT4,GOS)
Job Scheduler (OpenPBS,LSF,Conder)
Grid Resources(Cluster, expensive instrument)G
ang
lia
CVS ServerDatabase ServerMysql
IE browserarchitecture of CMAG Portal
Security and user management
• provides a single-log-on access interface for all grid services and grid resources
• authenticated users can access to computing, storage, data, software resources as well as all kinds of services
Security and user management
• SimpleCA of Globus toolkit is adopted to ensure the security of the grid resources– based on GSI and supports X.509 inter-certification mechanism
• CMAG Certification Authority signs each certificate
• In actual implementation, a CA center at CMA is set up and CA distribution packages are installed at all CMAG nodes or provincial meteorological bureaus
• User certifications managed by Myproxy library
CMAG Portal• Resource Access Portal
– to access geographically distributed computing resources
• NWP Products Browse Portal – to disseminate NWP products to the public and professional clients
• Remote Collaborative Portal (RCP)– to offer a universal platform for scientists to develop NWP modules, to share
source codes, to register modules within the grid and to perform NWP experiments.
CMAG Portal
• Globus 4.0 as the communication and tool-kit for collaborative research
• authentification and grid-job management is performed based on Gridsphere Portal Framework and Java COG tool-kits.
Infrastructure of CMAG Portal
Job Submission and resource monitoring
• based on the pre-wsrf-gram and wsrf-gram of Globus 4.0
• Each grid resource has its own job scheduling software:– SSC:
• PLATFORM’s LSF to perform job scheduling
– Others:• OpenPBS
• Ganglia to display dynamically the load of computing resource :– such as CPU load, availability of both memory and ha
rd-disk, as well as network throughput
The load report of grid resources displayed dynamically by Ganglia within CMAG
NWP executing control interface ---
GridWeather
• NWP system involves complicated process control and massive parameters setting
• traditional implementation method:– write a control routine – effective in local mode familiar with both every part of
NWP system and system environment.
• Gridweather:– based on Portlet including GRAM and GridFTP of Co
G as well as Globus toolkit
interface for research: GARPES model improvement
– to select the appropriate physical parameterization schemes
– to designate forecast domain, available numerical schemes, etc.
Flowchart of GRAPES model improvement on CMAG
job-submission strategy : firstly, select the computing resources already with special experimental data, secondly, select machines that can run the GRAPES model
NWP operational executing interface
Flowchart of operational GRAPES meso-scale model on CAMG
job-submitting strategy:
(1) selecting the computing resources that have had data already
(2) selecting idle machines, then local machine, or any available high-performance computing capabilities machines.
CVSExplorer
• NWP system upgrading• frequent modifications• diversity of program versions• intensive interactions between model devel
opers and inter-comparisons between numerical experiments
• it is necessary to create a suitable platform for joint NWP software development based on WinCVS
• Browse source codes
• GRAPES improvement
Configuration of CMAG
• portal server:– deployed on a DELL server at CAMS, includes:
• Web server Tomcat• Portal Framework Gridsphere• Web service software Java SOAP• grid application and development package CoG, GMETAD• web Front-end Ganglia
• management node:– implemented at OSCAR Cluster at CAMS
• Hierarchically structured MDS is installed– GIIS is available at each managing node and GRIS at
other nodes.
Computing resources
Location Grid Software
Security NWP software Job scheduler
Oscar CAMS GT4 CMA CA Center
GRAPES OpenPBS
Sunway48I CMAS GT4 CADP GRAPES OpenPBS
Sunway32I NMIC GT4 CADP GRAPES OpenPBS
Sunway32P NMIC GT4 CADP GRAPES OpenPBS
Ocean STI GT4 CADP GRAPES OpenPBS
Dawning GRMC GT4 CADP GRAPES OpenPBS
Dawn GXMB GT4 CADP GRAPES OpenPBS
PC Wuhan GT4 CADP GRAPES No
Dawning 4000A
SSC GT3 CADP GRAPES LSF
CADP: CMA CA Distributed Packages
STI: Shanghai Typhoon Research Inst.
Computing platform on CMAG
Real-time weather forecasts
• GRAPES meso-scale model 30KM/15KM– Mainly for rain forecast– runs every day on CMAG
• Real-time forecast products:– http://grid.cma.gov.cn:8080/
• Also computing resources and NWP technique comparatively less developed regions can utilize the grid resources or even run the NWP models to satisfy their forecast service requirement:– Yantai– Qinghai
GRAPES-Meso Scale model
• new NWP system developed by Research Center for Numerical Prediction at CAMS
• Features:– fully compressible primitive equations– switchable between hydrostatic and non-hydrostatic mode– semi-implicit and semi-Lagrangian advection schemes – latitude and longitude grid– horizontal Arakawa C grid – terrain-following coordinates – Charney-Phillips staggering vertically– physical parameterization package
ATOVS Data
Preprocess
Quality control
ConventionalData
preprocess
Quality control
3DVariation Anal
ysis
3DVariation Anal
ysis
Global model lateral B.C.
GRAPES Model real-timeforecasts
GRAPES Model real-timeforecasts
Post-processing
Post-processing
Forecast products
visualizationvisualization
prepocessing
Initial 48hrs forecast
6hrs cycling
Analysis Forecast Postprocessing
Flowchart of meso-scale GRAPES model on CMAG
500 hPa Geopotential height forecast over China (2005090900UTC)
forecast
Hourly rain forecast for Beijing, 2005080800UTC
Typhoon MTyphoon Matsaatsa
Conclusion
• Distributed computing resources at Beijing, Shanghai, Guangdong and Guangxi aggregated
• real-time GRAPES executions on CMAG feasible
• a virtual collaborative R/D center established:– provides researchers at different locations with a
common R/D environment aimed at further improvement of the GRAPES
Thanks for listening!