High-Energy Physics Applications on the Grid - Grid Computing: Infrastructure, Service, and Applications - page 426

17.3
The LHC started operation in 2008. Four major LHC experiments (ALICE,
ATLAS, CMS, and LHCb) will collect roughly 15 PB of data per year, which
should be processed, calibrated, and analyzed multiple times. Seamless
access to the LHC data should be provided for 5000 scientists in about 500
research institutions worldwide. The lifetime of the project is estimated to
be around 20 years.
The goal of the LCG (also called Worldwide LCG or WLCG) is to pre-
pare and deploy the computing environment indispensable to perform
the physics program of the LHC project. This includes the detector
simulation studies to push the detectors' performance to their limit, the
calibration and the monitoring of the detectors during data taking, the
reconstruction of the raw data, and other selection and analysis stages. All
relevant data should be accessible to all the physicists worldwide partici-
pating in an experiment.
The LCG Technical Design Report [10] estimates the computing power
required for the LHC data analysis to be of the order of 100,000 CPUs (CPU
available in 2004). A globally distributed model for data storage and analy-
sis was chosen. Originally the MONARC project (Models of Networked
Analysis at Regional Centers for LHC experiments) suggested a hierarchical
structure of distributed computing resources (partially modii ed due to the
emerging grid technologies). CERN and multiple computing centers world-
wide are providing resources for constructing the LCG infrastructure.
The infrastructure, which has been built, has a hierarchy of tiers of
computing centers. CERN is the Tier0 center of the infrastructure. The
primary functions are the data recording and the permanent storage capa-
bi l it y (t ap e s y st e m). T he s y st e m shou ld b e c apable to su st a i n up to 1. 2 5 GB/s
of data recording rate (ALICE experiment during heavy-ion operations)
and store several tens of PB per year. Tier0 provides CPU power for data
calibration and i rst-pass reconstruction. Tier0 distributes data to the Tier1
according to the policies agreed with each experiment.
The infrastructure has 11 Tier1's. Each Tier1 has custodial responsibility
for the data received from Tier0 and for data processed in the Tier1 layer.
Tier1 CPU will be used heavily in data reprocessing and in preparing big
data sample for analysis. The Tier1's are: ASGC Taipei, BNL US, CNAF-
INFN Italy, FNAL US, GridKa Germany, IN2P3 France, NDGF in the Nordic
countries, NIKHEF/SARA Netherlands, PIC Spain, RAL UK, and TRIUMF
in Canada. All Tier1s have support and data distribution (and custodial)
responsibility to the next level in the hierarchy, the Tier2 centers.
So far, over 100 Tier2s are participating in LCG. At variance with the Tier0
and Tier1s, Tier2s have no long-term data storage responsibility. Ultimately
they will provide the computing resources for most of the analysis activities
(hence serve the majority of the users). Tier2s also have a very important

The LHC and the Grid Projects