The data should
include, for each experiment:
- the network
bandwidth implications for each Tier-1/Tier-2 pair
- the data storage implications for each
Tier-1 as a function of the associated
Tier-2s
All known resources
should be included, with the status of each site being
indicated (e.g. pledged in MoU, expected to be included in the MoU,
resources presently outside the
collaboration).
The aim is
to allow sites and experiments to check that there is no
over-commitment of storage resources, and
allow a first assessment of possible network bandwidth issues to be made. This may well give rise to requests
from sites to make changes to ease support and operational relationships, but
after a few iterations we should be able to reach a reasonably stable situation
- with undoubtedly some problem cases. Sites will then have a good understanding of the
data links that they have to test, the characteristics of the FTS channels that
they will have to manage, and the other sites
with which they will have to operate.
The team will consist
of one person from each experiment, preferably someone who
has been involved in the the preparation of the data presented recently to the GDB and OB, coordinated
by the LCG Resource
Manager, Chris Eck, who will maintain a centralised copy of the
data. Once a first version of the data has
been assembled the team should organise reviews by sites and network experts,
and oversee a number of iterations until the data is reasonably stable.
Progress and any problems and difficulties identified should be reported
regularly to the MB. A summary must be prepared for the OB meeting on 11
September.
It is envisaged
that the work of the team would be completed in 4-6 months, after which the data
would be maintained by a straightforward reporting mechanism (cf. the Phase 2
Planning Group and the MoU resource
tables).
____________________________________________
a team
consisting of Nick Brook (LHCb), Chris Eck (LCG), Roger Jones (ATLAS),
Dave Newbold (CMS), Yves Schutz (ALICE) has prepared a first version of the
requested data.
The workbook attached to this mail contains for each WLCG
Tier-1 centre a table listing for all Tier-2 centres or federations supported by
this Tier-1 centre the amount of storage required at the Tier-1 centre and the
network bandwidth required to ship data from the Tier-1 to each connected Tier-2
and similarly the bandwidth required for sending data created at the Tier-2s to
the storage at this Tier-1.
The required storage capacity is split into
the three WLCG storage classes (tape only, tape and disk, disk only). As the the
Tier-2 requirements at the Tier-1 are dependent on the computing models of the
experiments and are calculated and provided by a representative of each
experiment, these requirements of the Tier-2 centres or federations are split
according to the served experiment and the table shows also for each experiment
the CPU and disk capacity pledged to the served experiments by each Tier-2 in
the reference year 2008.
The preparation of the WLCG MoU pledge tables
was done sofar under the assumption that 2008 was the first full running year of
the LHC. This will certainly not be anymore the case, but we have until now
neither a new ramp-up schedule for the required computing capacities from the
experiments nor adapted pledge tables from the Tier-1 and Tier-2
centres/federations. The team preparing the attached tables decided therefore to
base the first version of these tables on the existing data for "the first full
year" of LHC running. We believe that the resulting requirements will still be
helpful for resource planning in the Tier-1 and Tier-2 centres of the
WLCG.
A few further comments need to be made concerning the Tier-2 pledge
tables in the attached workbook. The Tier-2 centres/federations are sorted into
three groups, separated by a blank row. The first group contains Tier-2s already
in the WLCG MoU with their corresponding pledges. The second group contains the
centres, which have already been announced to the C-RRB as likely candidates
planning to join the WLCG. The third group contains Tier-2s seen by the
concerned experiments as additional candidates and added in the tables to give
the most complete picture possible at this moment. Resource pledges entered for
the last two groups have certainly to be verified by the concerned
sites.
Another point to note concerns the bandwidth and storage
requirement entries for ALICE. Tier-2 pledges for ALICE in the WLCG MoU cover at
this moment only ~50% of ALICE's requirements expressed in its Computing TDR. As
ALICE is still expecting major increases in these pledges, all requirement
figures for ALICE in the first group of Tier-2s (the one covered by the MoU) are
based on pledged Tier-2 computing capacities which are linearly increased by a
factor two above the pledge figures in the tables. In addition, ALICE has added
a further Tier-1 to the tables which is also not yet in the MoU.
The
other experiments base their requirements calculation on the MoU pledges and the
best available figures for Tier-2s of the last two groups. Still, CMS wants to
point out that it expects a major increase in the RAL pledges for CMS which will
lead to revised requirements figures for this Tier-1. A point which is
concerning all experiments, but has been made most strongly by ATLAS, concerns
the "balance" of CPU to disk capacity in the Tier-2s. There exists a good number
of cases where a better alignment of the ratio CPU/disk to the TDR requirements
of the experiments would result in better computing efficiency for the same
investment at the Tier-2.
Finally a few remarks about efficiency factors.
CPU efficiencies of 85% for well behaved calculations like simulations and 60%
for "chaotic" analysis have been included in calculating the Tier-2
requirements. To the contrary, the bandwidth figures in the table are just
arrived by dividing the amount of data to be transported by the time available
for doing it. It is left to the network experts to multiply these net bandwidths
with some factor of their choice before comparing them with the raw bandwidth
capacities of the available networks.
Storage requirements in the tables are
also net values. The Computing TDRs assume 100% efficiency for tape storage and
70% for disk storage. To help in assessing the part of the pledged disk capacity
at each Tier-1 required for the served Tier-2s the total amount of disk used at
70% efficiency for the Tier-2s is calculated at the bottom of each table and
compared with the disk pledge for 2008 in the MoU. Please note that the disk
capacity required as cache in front of the tape only storage is not included in
this calculation.
Please take the time to verify that the requirements
listed in these tables are compatible with your resource planning. If in some
cases this should not be the case, possible adjustments will obviously need to
be discussed between the concerned experiments and the Tier-1/2
centres/federations. Yet, it would be helpful, if the LCG Resource Manager could
be informed about the existence of these problems.
For the Tier-2/Tier-1
Team,
Chris
Christoph Eck
LHC Computing Grid Project
Resource
Manager
Information Technology Department
European Organization for
Nuclear
Research
CERN
phone: +41 22 7674260
CH-1211 Geneva 23 gsm: +41 76
4873800