Upload
darrius-echo
View
18
Download
2
Embed Size (px)
DESCRIPTION
CMS GRID COMPUTING AT THE SPANISH TIER-1 AND TIER-2 SITES. For more information, contact: Jos é M. Hernández , CIEMAT, Particle Physics Division E-28040, Madrid, Spain [email protected]. - PowerPoint PPT Presentation
Citation preview
Tier-2 Data Analysis MC simulation Import data from Tier-1 and export MC data
CMS GRID COMPUTING AT THE SPANISH TIER-1 AND TIER-2 SITES
P. Garcia-Abia a, J.M. Hernández a, F. Martínez b, G. Merino b, M. Rodríguez b, F.J. Rodríguez-Calonge a a CIEMAT, Madrid, Spain b PIC, Barcelona, Spain
Abstract
CMS has chosen to adopt a distributed model for all computing in order to cope with the requirements on computing and storage resources needed for the processing and analysis of the huge amount of data the experiment will be providing from LHC startup. The architecture is based on a tier-organized structure of computing resources, based on a Tier-0 center at CERN, a small number of Tier-1centers for mass data processing, and a relatively large number of Tier-2 centers where physics analysis will be performed. The distributed resources are inter-connected using high-speed networks and are operated by means of Grid toolkits and services. We present in this poster, using the Spanish Tier-1 (PIC) and Tier-2 (federated CIEMAT-IFCA) centers as examples, the organization of the computing resource in CMS together with the CMS Grid Services, built on top of generic Grid Services, required to operate the resources and carry out the CMS workflows. The Spanish sites contribute with 5% of the CMS computing resources. We also present the current Grid-related computing activities performed at the CMS computing sites, like high-throughput and reliable data distribution, distributed Monte Carlo production and distributed data analysis, where the Spanish sites have traditionally played a leading role in development,integration and operation.
CMS Computing Model
Tier-0 Accepts data from DAQ Prompt reconstruction Archives and distributes data to Tier-1’s
Tier-1 Real data archiving Re-processing Calibration Skimming and other data- intensive analysis tasks MC data archiving
DataTransfer DB
TMDB
Data Book-keeping DB
RefDB
LCG replicacatalogue
User Interface
MC Production McRunjob
Analysis ToolCRAB
Job Monitoring& Bookkeeping
BOSS
Data Location DB
PubDB
ResourceBroker
Grid InformationSystem
Computing Element
WorkerNode
StorageElement
Data TransferPhEDEx
Site LocalCatalogue
Application
Database
LocalGlobal Local or remote
Computingresource
Interaction
Job flow
Data flow
Wor
kloa
d M
anag
emen
t S
yste
mD
ata
Man
agem
ent
Sys
tem
CMS Computing Services/Workflows
End 2005
End 2006End
2007
CPU (kSI2k) 100 300 800
Disk (TB) 50 200 400
Tape (TB) 100 400 800
WAN (Gbps) 1 10 10
End 2005
End 2006End
2007
CPU (kSI2k)
150 400 800
Disk (TB) 20 100 200
Tape (TB) - - -
WAN (Gbps)
1 1 10
Hardware Resources
PIC Tier-1 Tier-2 SpainCIEMAT-IFCA
Data Distribution
Computing Grid Activities
• Done with PhEDEx, the CMS large scale dataset replica management system. Managed, reliable, routed multi-hop transfers. Includes pre-staging data from tape• Demonstrated sustained transfer rates of ~3TB/day to the Spanish sites.
Distributed Data Analysis• Spanish sites host ~15% of CMS data (~15 Mio. events, ~15 TB)• ~10K analysis jobs ran at the Spanish sites so far
Distributed Monte Carlo Production• Leading role of the Spanish sites in MC production in LCG Grid in the areas of development, integration and operation• ~ 15 Mio. MC events produced in LCG
Production activities
LCG Service Challenge 3 CMS goal: exercise under realistic conditions bulk data transfer and data serving infrastructure
SC3 Throughput phase (July 2005) Goal: high throughput transfer + storage system test Sustained rate disk-to disk of ~4TB/day (50 MB/s avg) for 2 weeks to PIC Tier-1
SC3 Service phase (September-November 2005) Goal: concurrent data transfer and data access from storage Test Data Management components (data transfer system, data publication, data catalogues, storage systems) as well as Workload Management components (job submission, monitoring, etc) in a large-scale real use. Automatic data publication for analysis as data get transferred Automatic analysis job submission as data get published 15 TB of data transferred to the Spanish sites 10K jobs run at the Spanish sites (~20% of the CMS SC3 jobs run in LCG) 175 MB/s aggregate analysis job read throughput at PIC Tier-1 and 75 MB/s at Spain Tier-2
Integration/test activities
For more information, contact:José M. Hernández,CIEMAT, Particle Physics DivisionE-28040, Madrid, [email protected]