14 Distributed Computing
635
Table 14.1 Capacity
deployed in the tiers of the
LHC computing grid in 2018
2018
CPU (KHS06 a )
Tier-0 b
1272
All Tier-1
2061
All Tier-2
2562
Disk (TeraBytes)
Tier-0
90700
All Tier-1
192214
All Tier-2
185455
Tape (TeraBytes)
Tier-0
284700
All Tier-1
461050
a Thousands of HEPSPEC06 units. HEPSPEC06 is
a measure of CPU speed maintained by the particle physics community. See http://w3.hepix.org/
benchmarking.html. As a very rough guideline, a
current 2 GHz CPU delivers about 15 HS06
b In addition to the Tier-0, CERN also deploys an
Analysis Facility (CAF) which is not included in
this table
coherent distributed system which supports all data processing operations of the
LHC experiments. WLCG defines a global service supported in part by Grid
Infrastructure services in various countries and regions, for example EGI in Europe
and OSG in the USA. The large amount of data handled by WLCG (over 100
Petabytes per year exchanged between sites) requires the sites to be organized in
three “Tiers”. Centers belonging to a given Tier are dedicated to perform certain
specific tasks within the computing model of a given LHC experiment. There is
one Tier-0 center at CERN, which handles Data Recording, Prompt Reconstruction
and Remote Data Distribution (mostly to the Tier-1 centers). There are 13 Tier-1
centers, which perform Data Recording (between them they hold a second copy
of all raw data stored at the Tier-0), Reprocessing, Data Reduction and Event
Filtering. Hundreds of Tier-2 centers receive from the Tier-1 centers DST-type files
(reduced and filtered) which physicists then use for analysis. Tier-2 centers also run
event simulation, transmitting the outputs to Tier-1 centers for data recording and
reconstruction. FTS, coupled to some experiment specific tools, is used to automate
all data transmission. 75 The capacities deployed are quite large: about 400 thousand
computing cores, 250 Petabytes of disk and nearly an Exabyte of tape. Table 14.1
shows the capacities deployed in 2018. These capacities have grown at an impressive
rate of about 37% year-to-year since 2008. An even larger growth is expected for
Run-3 of the LHC starting in 2021.
75 The Wide Area Network requirements between the Tier-0 center and the 13 Tier-1 centers are
so demanding that a special network based on point-to-point 10 Gbps links has been deployed
(the LHC Optical Private Network, or LHCOPN). As of this writing, the LHCOPN links are being
upgraded to 100 Gbps.
635
Table 14.1 Capacity
deployed in the tiers of the
LHC computing grid in 2018
2018
CPU (KHS06 a )
Tier-0 b
1272
All Tier-1
2061
All Tier-2
2562
Disk (TeraBytes)
Tier-0
90700
All Tier-1
192214
All Tier-2
185455
Tape (TeraBytes)
Tier-0
284700
All Tier-1
461050
a Thousands of HEPSPEC06 units. HEPSPEC06 is
a measure of CPU speed maintained by the particle physics community. See http://w3.hepix.org/
benchmarking.html. As a very rough guideline, a
current 2 GHz CPU delivers about 15 HS06
b In addition to the Tier-0, CERN also deploys an
Analysis Facility (CAF) which is not included in
this table
coherent distributed system which supports all data processing operations of the
LHC experiments. WLCG defines a global service supported in part by Grid
Infrastructure services in various countries and regions, for example EGI in Europe
and OSG in the USA. The large amount of data handled by WLCG (over 100
Petabytes per year exchanged between sites) requires the sites to be organized in
three “Tiers”. Centers belonging to a given Tier are dedicated to perform certain
specific tasks within the computing model of a given LHC experiment. There is
one Tier-0 center at CERN, which handles Data Recording, Prompt Reconstruction
and Remote Data Distribution (mostly to the Tier-1 centers). There are 13 Tier-1
centers, which perform Data Recording (between them they hold a second copy
of all raw data stored at the Tier-0), Reprocessing, Data Reduction and Event
Filtering. Hundreds of Tier-2 centers receive from the Tier-1 centers DST-type files
(reduced and filtered) which physicists then use for analysis. Tier-2 centers also run
event simulation, transmitting the outputs to Tier-1 centers for data recording and
reconstruction. FTS, coupled to some experiment specific tools, is used to automate
all data transmission. 75 The capacities deployed are quite large: about 400 thousand
computing cores, 250 Petabytes of disk and nearly an Exabyte of tape. Table 14.1
shows the capacities deployed in 2018. These capacities have grown at an impressive
rate of about 37% year-to-year since 2008. An even larger growth is expected for
Run-3 of the LHC starting in 2021.
75 The Wide Area Network requirements between the Tier-0 center and the 13 Tier-1 centers are
so demanding that a special network based on point-to-point 10 Gbps links has been deployed
(the LHC Optical Private Network, or LHCOPN). As of this writing, the LHCOPN links are being
upgraded to 100 Gbps.