29 may 2002joint edg/wp8-edt/wp4 meetingclaudio grandi infn bologna lhc experiments grid integration...
TRANSCRIPT
29 May 2002Joint EDG/WP8-EDT/WP4 MeetingClaudio Grandi INFN Bologna
LHC ExperimentsGrid Integration Plans
C.Grandi
INFN - Bologna
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 2Claudio Grandi INFN Bologna
Introduction
What do LHC experiments have to do:Build a distributed computing system for the experiment
Test the prototypes:tools evaluation
data challenges: tests with big data flows
What do LHC experiments have today:Software
• EDG release 1.1 (1.2 coming)
• VDT release 1.0 (includes Globus, Condor-G, GDMP, …)
• Other tools for distributed computing (e.g. Web services, …)
Test facilities• EDG test bed
• EDT supported experiment test beds
• US GriPhyN/PPDG/iVDGl experiment test beds
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 3Claudio Grandi INFN Bologna
ALICE: Overview
Application: AliRoot
Data Persistency: RootJob parallelisation: Root/PROOF
Job submission/control/monitoring: AliEn
Data Catalogue: AliEn
AliRoot AliEn API/PROOF
AliEn Services
ROOT
IVDGL EDT
By P.Cerello
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 4Claudio Grandi INFN Bologna
ALICE: Sites
Houston Univ. - TexasOSU/OSC - OhioLBL - California
CERN - CHINFN Cagliari/Catania/ CNAF/Salerno/Torino - INIKHEF - NL (?)
By P.Cerello
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 5Claudio Grandi INFN Bologna
ALICE: Integration Items
Job submission: EDG/Resource Broker as an AliEn client Condor-G as an AliEn client
Data Management: register/access the AliEn Data Catalogue from an
EDG/iVDGL/... Job
MetaData Catalogue: implement a Job MetaData Catalogue Test multiple accesses and concurrent updates Evaluate Spitfire to manage it
Spawn PROOF sub-jobs on any GRID
By P.Cerello
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 6Claudio Grandi INFN Bologna
ALICE: Integration Plans
AliEn Job
AliEn Job
AliEn Job
AliEn Job
AliEn Job
AliEn Job
AliEn Job
AliEn Job
EDG RB
EDG CE
EDG WN
AliEn DataCatalogue
EDG SE
Condor-G
Alice Farm
Alice User
I-LFN
I-PFN
O-PFN
O-LFN,O-PFN
MetaDataCatalogue
By P.Cerello
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 7Claudio Grandi INFN Bologna
ALICE: Integration Plans
EDG WN
AliEn DataCatalogue
EDG SE1
I-LFN(s)
O-PFN3
I-PFN(s)
I-PFN3
MetaDataCatalogue
I-LFN(s)
EDG SE3
EDG SE2
EDG CE2
EDG CE3
I-PFN2
I-PFN1
O-PFN2O-PFN1
PROOF
By P.Cerello
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 8Claudio Grandi INFN Bologna
ATLAS: DC1 & 2
DC1/1: simulation of 10**7 evts for HLT TDR started 15 April. Bulk of production will start after 15 June
DC1/2: stress is on new software (20-9 till Xmas)
Use of currently avalable Grid tools favoured but not mandatory: start with EDG 1.2 in EU and VDT 1.0 in US:Some US sites deploy EDG software and vice-versa.
Integration of Magda (US data catalog+some replica management) with EDG RC
Critical issue the EDG stability and reliability
DC2: start spring 03Use EDG release 2
Grid API to be inserted in ATHENA framework
Magda and WP2 convergency plan to be detailed and executed
Integrate VDT components
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 9Claudio Grandi INFN Bologna
ATLAS: Overview
Main tools for data management on the grid:
Magda - MAnager for Grid Data
GANGA - Gaudi and Grid Alliance (with LHCb)
Other tools in development:– GridView - simple script tool to monitor status of
testbed (Java version being developed)– Gripe - unified user accounts– Pacman - package management and distribution tool– Grappa - web portal based on active notebook tech.– GRAT (GRid Application Toolkit): distribution of
ATLAS software on the Grid
Integration with EDG is being discussed
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 10Claudio Grandi INFN Bologna
ATLAS: Magda
Magda:Magda: MA MAnager for GGrid-based DDataDeliverable for the PParticle PPhysics DData GGridMySQL database at the core of the systemUser interaction via command line and web interfaceweb interface File replication via GSIFtpCurrently in use for ATLAS distributed data
management prototyping and design
260k files, ~11TB cataloged at present
Synchronization between Magda & EDG cataloguesATLAS applications query Magda’s MySQL catalogue;EDG Resource Broker queries the Globus Replica
Catalog.By D.Rebatto
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 11Claudio Grandi INFN Bologna
ATLAS: Magda
Magda
ExportCatalogue
ImportCatalogue
LocalCatalogue
Magda
ExportCatalogue
ImportCatalogue
LocalCatalogue
Host A Host B
GDMP GDMP
Globusreplicacatalog
MagdaMySQLcatalog
USER B: gdmp_host_subscribe -r host_b -p <p>
my_file
A file named my_file is created at Host A
LocalCatalogue
my_file
USER A: gdmp_register_local_file -p my_file
GDMP
LocalCatalogue
my_file
USER B: gdmp_replicate_get
GDMPGSIFtp
my_file
Magda
LocalCatalogue
my_file
GDMP B: magda_putfile my_file --registeronly
GDMPGSIFtp
my_file
GDMP_STAGE_TO_MSS
GDMP
ExportCatalogue
ImportCatalogue
my_file
USER A: gdmp_publish_catalogue
GDMP
Magda
GDMP
ExportCatalogue
ImportCatalogue
my_file
GDMP A: magda_putfile my_file --registeronly
GDMP
By D.Rebatto
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 12Claudio Grandi INFN Bologna
CMS: Overview
Build a unique CMS-GRID framework (EU+US)
EU and US grids not interoperable todayHelp from DataTAG-iVDGL-GLUE
Work in parallel in EU and US
Main US activities: MOP
Virtual Data System
Interactive Analysis
Main EU activities:Integration of IMPALA with EDG WP1+WP2 sw.
Batch Analysis: user job submission & analysis farm
CMS Production Framework
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 13Claudio Grandi INFN Bologna
CMS: Schedule2000-01: A few sites in production. Some grid
tools used “in production” (GDMP).2002: World-Wide productions. Prototypes of
“distributed” production sites deployed in US and EU. DAQ TDR.
2003: Integration activities: US/EU grid interoperability. First grid-enabled sites used “in production”.
2004: 5% data challenge. Deployment of delivered LCG prototypes. CCS TDR.
2005: preparation for 20% data challenge. Build "final" computing environment. Physics TDR.
Long term schedule being revised now!
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 14Claudio Grandi INFN Bologna
Data Management components-Replica Manager-Virtual Data Catalogue-Replica Catalogue-…
Planner (uses CMS
production SW)
Storage Element
Reference DB
Local analysis tools with plugin to access
remote data
Physics Data Production & Analysis Portal
Computing Element
EDG SW
= data flow or invocation = read data from = contains reference to
= active component = catalog or database
job job
job
job
job
CMS: EU-US integration 02
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 15Claudio Grandi INFN Bologna
Computing Element
Local Scheduler
GRAM
Replica Catalogue
Maps each logical file name toone or more physical file names
Job Submission Service
Resource BrokerFinds suitable
Location for execution
Condor-G
Worker nodes
CMSIMORCA
Storage Element
Local StorageLocal ObjyFDDB
Reference DB
has all informationneeded by IMPALA
to generate a dataset
User Interface
job
job
= data flow or invocation = read data from = contains reference to
= active component = catalog or database
Information Services
LDAP serverResource
information
job
Production portal
IMPALA
Get request for a productionCreate location independent jobs
BOSS
CMS: EDG integration 02
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 16Claudio Grandi INFN Bologna
LHCb: Overview
GANGA (Gaudi and Grid Alliance) :In collaboration with Atlas developing user interface for
all levels of user (physicist,production manager,developer)
Control and Monitoring system for distributed data production:Using PVSS today
Data challenges:DC1: 3-30 june + 26 august -22 september 2002
DC2: 2003
DC3: 2004
DC4: 2005
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 17Claudio Grandi INFN Bologna
LHCb: DC1Computing tests
– prototype of new data management databases– configuration of LHCb environment at remote sites and
installation kits– monitoring and control with PVSS– integrability of EDG testbed into our production system– data quality checking, evaluate DaVinci based tools– use "push" MC job submission initially, perhaps test first
version of pull mechanism during second half– executables to be tested in production: sicbmc, Brunel, DaVinci
Data productionProduce useful data, for example for Physics studies with this
facility, if the software is ready…
Resources:100 – 150 CPU at CERN & ~300-350 CPUs outside (Bologna, IN2P3,
RAL, Oxford, Bristol, Edinburgh, Nikhef, Cambridge, Barcelona, Moscow, Amsterdam VU?) By J.Closier
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 18Claudio Grandi INFN Bologna
LHCb: DC2 (2003)Computing tests
– production version of new data management databases– implementation of "pull" MC job submission system, exploiting
EDG resource broker– stress test Grid "philosophy" of submitting jobs without worrying
where the job will run and where the output will be stored– monitoring and control, evaluate EDG middleware (how does it
work with PVSS? can it replace it?)– make job submission tools independent of remote sites– test of first Ganga prototype– test of some Gaudi-Grid interfaces e.g. event selector– test of automatic data quality histogram creation and checking in
production– executables to be tested: (sicbmc), Gauss, Brunel, DaVinci, Gaudi
ResourcesSame sites as DC1 + Switzerland, Germany, Poland
By J.Closier
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 19Claudio Grandi INFN Bologna
LHCb: DC3&4 (2004/5)
Computing tests for DC3– test of first prototype of the analysis model (display
events that were analysed on the grid)– test of Ganga production version– full use of Grid in production– further tests of Gauss in production– tests of data quality checking in production– executables to be tested: Gauss, Brunel, DaVinci,
Gaudi
Computing tests for DC4test of production analysis model
the other points also mentioned previouslyBy J.Closier
29 May 2002Joint EDG/WP8-EDT/WP4 Meeting 20Claudio Grandi INFN Bologna
Summary
Experiments are integrating Grid tools in their computing environmentMost of the experiments are using Grid tools delivered
by more than one Grid Project
Most of the experiments are not planning to use the tools out of the box but to adapt them to their needs
Need help by middleware experts for interoperability
Need non-standard facilities to develop and test the experiment computing environment. Help by middleware experts is needed for that too