ITPA/IMAGE conference 7-10 May 2007

ITPA/IMAGE 7-10 May 2007

Software and Hardware Infrastructure for the ITM

B.Guillerminet, on behalf of the ITM & ISIP teams

(P Strand, F Imbeaux, G Huysmans, J Lister, G Manduchi, F Iannone, M Airaj, P Huynh …)

and others(Y Smirani, V Bergeaud, R Metery, D Drouin, P Micoud, C Leroy …)

ITPA/IMAGE conference7-10 May 2007


Software & hardware infrastructure for the ITM

• ISIP belongs to the Integrated Tokamak Modelling project• In charge of the infrastructure (soft + hard) for the ITM:

– Code platform: framework, editor, …

– Data handling: data structure, data storage & access, …

– Tools: version handling, portal, …

– Hardware: gateway, GRID/HPC, …

• 3 others talks: – ITM (P Strand)

– Data structure (F Imbeaux)

– Scientific workflow with KEPLER (M Airaj)

• Focus on the software architecture


Outline• Requirements & challenges

• Choices & solutions

• Architecture

• Data structure

• Data access & storage: UAL

• Code structure

• Intensive computing: GRID/HPC




• Architecture

• Data structure


• Code structure



Requirements & challenges

• Long lasting project: – O.S. will change– New languages– Framework could disappear

• Legacy software and language mixing:– Codes written in F77, F95, 2003, C, C++,

Java, Matlab, …– Depends on the Fortran compiler

• Local/remote computing:– Scheduling (start/suspend/resume/abort …)– Naming, discovery tools

• Fast communication:– how to deal with all the above (big/little

endian, language mixing, …)

Component:• Standard?• CCA?

BABEL

XML:• Binary

XML• Fast XML




• Architecture

• Data structure


• Code structure



Choices & solutions

• Architecture in layers

• Not an universal solution for any language to any language but valid for any identified data: XML schemas

• Standard code description: WSDL

• Binary data transfer: UAL

• Remote computing: Web Services

• Intensive computing: GRID, DEISA




• Architecture

• Data structure


• Code structure



Architecture

• Gateway – Cluster 32-64 CPUs (16GB), 10-100TB, 10Gb/s– UI: access to GRID EGEE

• Web portal:– Single sign-on, authentication, security, user’s profiles– Access to ITM applications– Catalogues and simulation storage


Architecture (cont’d)

• Applications:– Workflow engine (KEPLER):

• running

– Simulation editor/director:• Writing the requirements• Delivered at 4Q07

– Data mining tool

– SCILAB: post-processing tool• Work in progress

– Control & monitoring tools:• Cluster tools• GRID tools

– On-line data visualization:• 2D/3D actors in KEPLER (VTK based)• Visit tool (gateway + GRID)




• Architecture

• Data structure


• Code structure



Automatic generation:

Xsd2f, …

Data structure• Not an universal solution for any to any code but valid for any

identified data: XML schemas

DATA “equilibrium”• Description:

Structure: • F95 version • 3816 lines• 9MB

No need to know XML


Data structure (cont’d)

• XML schemas

• Generate automatically the include files for C, Fortran, Java, Matlab, …

• Use it in your code: #include …, import …

• Fill the structure through the provided API (UAL):

xsd2f,xsd2cxsd2m

No XML knowledge needed




• Architecture

• Data structure


• Code structure



Data access: UAL

Access: ITM library (UAL):• euitmopen()• euitmget_equilibria() • euitmput_equilibria()• euitmclose()

• Hidden data storage:– MDS+– HDF5– Commercial database– Message oriented Middleware (Control Host)

• APIs for “any” language:– C/C++– Fortran 77,95,2003– Java– Matlab

• Fast communication– Binary– In-memory, pointer access


codesEFIT, …

UAL as a layer• Communication between codes through the UAL• In-memory data server whiteboard

– Up-to-date data– Time-stamped data

• Independent of the framework

data serverITM + private

In-memoryJET, …Exp. data

UAL

codesHELENA, …

codesMISHKA, …


Publish-subscribeserver

in-memorydata

server

Soloviev

Helena

Mishka

Remote web server

KEPLER

Based on a Message Middleware:ControlHost (A.Maslennikov)

“equilibrium” data~9MB

display

Advantages:•distributed•fast (binary data transfer)•Language agnostic argument

s of the WS

UAL: MoM implementation




• Architecture

• Data structure


• Code structure



Code Structure• Independent of the framework:

– Able to change it later (layer)– Tools to integrate it (one click option)

• Data access through the UAL:– Independent of the storage– The framework does not know our data

• Tools to integrate it in the framework:– Local (JNI based tool)– Remote (Web Service based tool)

• Structure:

prefire

fire()

postfire()

initialize()

Iterations

wrapUp()

No WS knowledge needed


program test-coderead file => fill incall code(in, out)read out => write out

endsubroutine code(data in, data

out)…compute…return

end

subroutine initialize()call euITMopen(‘…’)return

endsubroutine fire (variables)

call euITMget(‘type’, in)call code(in, out)call euITMput(‘type’, out)return(variables)

endsubroutine wrapup ()

call euITMclose(…)return

end

subroutine code(data in, data out)…compute…return

end

No change

KEPLER

Code Structure (cont’d)

UAL

Initial code


Based on KEPLER =>graphical tool=>advanced models of

computation =>directors=>hierarchical design

Framework: workflow engine

Run control buttons

Scheduler: time behavior

Catalogues


in-memoryData server

SOLOVIEV

HELENA

MISHKA

Plot 2DPlot 3D




• Architecture

• Data structure


• Code structure



GRID-HPC

• Gateway– Local codes (through UAL)– 64 CPUs

• Computers from the associations– Remote codes (through WS + UAL)

• GRID-EGEE Fusion V.O.– Distributed computing – 10K CPUs

• HPC-DEISA– Parallel computing– Network of super-computers

GLOBUS =>C/C++

library=>Java library

in development

Euforia??


GRID-EGEE Fusion V.O.

GLOBUS =>C/C++

library=>Java library

in development

GLITE =>Java library

in development

ITM =>Codes as WS=>UAL data

access

ProxyJob transfer via FTP

Data transfer via FTP

Job scheduling via JCL


Work plan

2007: development• KEPLER prototype: workflows, visualization 2D/3D, local/remote

codes (with graphical tools)• Web portal requirements• User interface requirements (Simulation editor)• UAL C/C++/Fortran APIs (MDS+)

• Gateway• User interface delivered (simulation editor)• Repositories (codes, Kepler, …)• GRID-EGEE Glite in Kepler

2008: production• Enhancements: Kepler, User interface • Data mining tools• Web portal

• GRID-EGEE ITM• GRID-DEISA

July 07

December 07

July 08

December 08

ITPA/IMAGE conference 7-10 May 2007

Documents

data storage access

data handling

identified data

ualremote computing

data equilibriumdescription

wsdlbinary data transfer

language mixing

itm applicationscatalogues