210 likes | 238 Views
RDMS CMS Computing Model. V.Ilyin, V.Gavrilov, O.Kodolova, V.Korenkov, E.Tikhonenko Meeting of Russia-CERN JWG on LHC computing CERN, October 10, 2005. RDMS CMS computing group (was approved by the RDMS CMS Collaboration Board in December, 2004)
E N D
RDMS CMS Computing Model V.Ilyin, V.Gavrilov, O.Kodolova, V.Korenkov, E.Tikhonenko Meeting of Russia-CERN JWG on LHC computing CERN, October 10, 2005
RDMS CMS computing group (was approved by the RDMS CMS Collaboration Board in December, 2004) representatives from all the RDMS CMS institutes RDMS CMS Computing Model Prepared on the base of detailed discussions and estimations Reported at RDMS CMS TF Meeting on June 16, 2005 at CERN
Conception: • Tier2 Cluster of institutional computing centers with a partial T1 functionality • summary resources at 1.5 level of the canonical Tier2 center for the CMS experiment + Mass Storage System • ~5-10% of RAW DATA, ESD/DST for AOD selections design and checking and AOD for analysis (depending on a concrete task) • Basic functions:analysis; simulation; users data support; calibration; reconstruction algorithm development … • Host Tier1 in LCG infrastructure: CERN RDMS CMS Advanced Tier2 Cluster Participating institutes: Moscow ITEP, SINP MSU, LPI RAS Moscow region JINR, IHEP, INR RAS St.Petersburg PNPI RAS Minsk (Belarus) NCPHEP Erevan (Armenia) ErPhI Kharkov (Ukraine) KIPT Tbilisi (Georgia) HEPI
RDMS CMS Advanced Tier2 Cluster (cont.)(as Tier2 kind centerwith some features of the Tier1 center). • to provide thefacilities for simulation tasks, a number of analysis tasks • and also fordetectors calibration, HLT and offline reconstruction algorithms and analysis tools development. • connectivity between the RDMS institutes: >1 Gbit/s Ethernet • computing farms at the institutes combined into GRID infrastructure • data for the each particular task located at the farm closer to the final user of the data. Processing of a part of raw datain addition to ordinary functions of Tier-2 (like maintaining and analysis of AODs, data simulation and user support) some part of RAW,ESD/DST and the proper set of AOD data will be transferred and kept at the RDMS CMS Tier2 cluster at the mass storage system (MSS), located in one or two institutes. It is needed both for a calibration/alignment of some CMS detector systems for which the RDMS is responsible and for creation and testing of reconstruction software applied to some particular physics channels.
RDMS CMS Advanced Tier2 Cluster (cont.)(resources/data sets/databases) - Some fraction of RDMS T2 (30% of resources) will be scheduled for MC simulation of the standard CMS MC samples including the detector simulation and the first path reconstruction. MC simulation will be distrubuted between the RDMS CMS institutes in accordance with the resources located at the institutes (the more resources, the more loading). The MC data will be moved to CERN or(and) stored locally. - As soon as RDMS T2 has some T1 functionality, it will get data both RAW, RECO (the amount needed for the software/trigger/calibration development) and AOD(for physics analysis of the dedicated channels). - RDMS T2 will publish all data located locally to be available for all CMS users, however the available datasets on the disks/tapes will reflect local user tasks. All CMS users will have an access to these data and will be able to submit the job through GRID. The non-local community will not be able to initiate the large-scale transfer to the site. - RDMS T2 will need the replication of databases (conditions), used in reconstruction.
RDMS CMS Advanced Tier2 Cluster (cont.)(canonical RDMS CMS user) A canonical RDMS CMS T2 user is considered to be a group of persons working on the dedicated task. A canonical user will participate in physics analysis and software/HLT/Calibration task for dedicated channel, i.e. needs AOD and a part of RAW/RECO data. A canonical user will submit 10-20 jobs daily (75% of jobs - to own grid resources and 25% of jobs - to the other common resources). The CPU needed is ~90 KSI2K. The mean value of TB/canonical user is ~30 TB in 2007. A number of canonical users is ~14 over all institutes
The estimates of resources for the RDMS CMS Tier2 cluster year by year: The estimates have been done in the accordance with the information provided by the RDMS CMS physics groups and the CMS Computing Resources Profile given in the CMS CTDR CPU Resources in MSI2K
The estimates of resources for the RDMS CMS Tier2 cluster year by year: Storage Resources in PB Data storage model is an open question Another point: resources sharing in a case when the groups from different institutes participate in the same physical task
The criteria of distribution of resources for CMS computing between Russian centres 1) existence of modern computing infrastructure with a proper hardware and software support including LCG environment and actual versions of CMS software: - high-speed communication channels; - equipped computer rooms with stable electric power supply and conditioning; - LCG–infrastructure; - technical staff providing twenty-four-hour operation of hardware/software complex 2) number of users - developers of algorithms; - developers of basic CMS software, data bases and data bases applications; - users on reconstruction, simulation, data analysis and data processing. 3)the existence of specialists responsible for installation, support and testing of actual versions of CMS software 4) level of participation in the physics analysis groups 5) financial investment of the institute into construction of CMS detectors .
Usage of CPU resources at Russian Tier2 during May-September, 2005 CMS jobs at Russian Tier2 sites in May-September, 2005: SINP MSU – 83 %, RRC KI - 11%, ITEP – 4%, JINR - 2%; IHEP, PNPI and INR sites – no CMS jobs during this period
CMS sw installed at RuTier2 LCG-2 sites IHEP:VO-cms-CMKIN_4_2_0_dar ITEP: VO-cms-CMKIN_4_1_0_dar;VO-cms-CMKIN_4_2_0_dar; VO-cms-CMKIN_4_4_0_dar; VO-cms-PU-mu_Hit3653_g133,VO-cms-OSCAR_3_6_5_SLC3_dar JINR: VO-cms-CMKIN_4_1_0_dar;VO-cms-CMKIN_4_2_0_dar; ; VO-cms-CMKIN_4_4_0_dar; VO-cms-OSCAR_3_6_5_SLC3_dar, VO-cms-ORCA_8_7_1_SLC3_dar RRC KI:VO-cms-CMKIN_4_2_0_dar; VO-cms-OSCAR_3_6_5_SLC3_dar; VO-cms-ORCA_8_7_1_SLC3_dar ; VO-cms-slc3_ia32_gcc323; VO-cms-ORCA_8_7_4 SINP MSU: VO-cms-CMKIN_4_2_0_dar; ; VO-cms-CMKIN_4_4_0_dar; VO-cms-OSCAR_3_6_5_SLC3_dar, VO-cms-ORCA_8_7_1_SLC3_dar, VO-cms-PU-mu_Hit3653_g133 PNPI:VO-cms-CMKIN_4_4_0_dar , VO-cms-ORCA_8_7_1_SLC3_dar INR:VO-cms-OSCAR_3_6_5_SLC3_dar, VO-cms-ORCA_8_7_1_SLC3_dar, VO-cms-slc3_ia32_gcc323, VO-cms-ORCA_8_10_1
RDMS CMS Data Bases (current status) ME1/1 Database: environment Database server: Oracle (provided by CERN IT and JINR LIT) WEB Server: Internet Information Server (Provided by CERN IT) ME1/1 Database: User Interfaces Web interface – provides the initial filling of database, different access levels for users, information search on different criterions, adding and updating data. HE Database: Tubes lengths measurements Radioactive source calibration. HF Database: Wedge calibration Beam wedge calibration Channel calibration. Databases & storage system structure
Dashboard Web UI Dashboard: CMS Job Monitoring R-GMA Monalisa RB RB WN CE R-GMA Client API Web Service Interface Collector (RGMA) Collector (Monalisa) Constantly retrieve job information Submission tools • Snapshot • Statistics • Job info. • Plots ASAP database PostgressqLite sqLite pg Others? RDMS CMS staff begins the participation in ARDA activities on monitoring for CMS PHP
http://www-asap.cern.ch/dashboard/ ASAP Arda Support for cms Analysis Process • Job Monitoring • (SC3 Jobs) • Running: 95 • Pending: 141 (updated time: 2005-10-06 18:35:40) • Transfer monitoring • Dynamic (Slower) • Historic (Faster) • Task Monitoring plans for monitoring development: monitoring of errors due grid-environment and automatic job restart in these cases; monitoring of a number of events done; the further expansion to private simulation (currently - for production&analysis)
Planning RDMS CMS participation in Service Challenge October-November session: RRC KI and SINP MSU SC04: RRC KI, SINP MSU and ITEP (?), IHEP(?), JINR(?) CMS data transfer system PHEDEX - already installed at SINP MSU and is in use The technical details of RDMS CMS participation in DCs should be determined in the nearest future at RDMS CMS sites and the practical details - in contact with collaboration at CERN
SUMMARY RDMS CMS computing group of representatives from all the RDMS CMS institutes was approved by the RDMS CMS Collaboration Board in December, 2004. The main current task of the group – to construct RDMS CMS Computing Model RDMS CMS Computing Model has been defined as Tier2 Cluster of institutional computing centers with a partial T1 functionality; the detailed description of the Model is presented The requirements on resources for RDMS CMS Tier2 Cluster determined in result of close discussions are presented The criteria of distribution of resources for CMS computing between Russian centres are formulated Current status of RDMS CMS Computing Support and the statistics on CPU usage at the LCG-2 Russian sites are presented ASAP to determine the details of RDMS CMS participation in SCs RDMS CMS DBs activities and participation in CMS Job Monitoring (in cooperation with ARDA) are presented in details.