Belle II computing model Belle II computing model (in relation to - PowerPoint PPT Presentation

Belle II computing model Belle II computing model (in relation to the EGI infrastructure) (in relation to the EGI infrastructure) Marko Bračko (Jožef Stefan Institute, Ljubljana & University of Maribor, Maribor) EGI Community Forum, EGI Community Forum, nd May 2014 Helsinki, Finland, 22 nd May 2014 Helsinki, Finland, 22

B- Factories (KEKB&PEP-II): B- Factories (KEKB&PEP-II): A Success Story A Success Story Quantitative confirmation of the KM model in the SM “… As late as 2001, the two particle detectors BaBar at Stanford, USA and Belle at Tsukuba, Japan, both detected broken symmetries independently of each other. The results were exactly as Kobayashi and Maskawa had predicted almost three decades earlier.” Marko Bračko DISCRETE`08, Valencia 2 M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 2

Introduction: Introduction: Physics Objectives of Belle/BaBar and Belle II Physics Objectives of Belle/BaBar and Belle II Discovery of CP violation in the B system Measurements of the CKM matrix elements PDG 2008 B 0 →J/ψK 0 _ B 0 →J/ψK 0 Asymmetry _ _ = (N-N)/(N+N) M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 3

Introduction: Introduction: Physics Objectives of Belle/BaBar and Belle II Physics Objectives of Belle/BaBar and Belle II ✔ Confirmation of KM B → D (*) τν mechanism of CP in the Standard Model ✗ CP in the SM too small (by many orders of magnitude) Measurement to generate observed matter-antimatter asymmetry in the universe Type II two-Higgs-doublet model (2HDM) ➔ Need sources arXiv:1205.5442 of CP beyond ➔ Super B factory the SM Complementary to the LHCb experiment at CERN M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 4

Introduction: Physics at a Super B Factory Introduction: Physics at a Super B Factory • What is the next experimental step? Precision measurements • Much larger sample needed for this purpose −> Super B factory • Hopefully new phenomena might be seen: – CPV in B decays from the physics outside the KM scheme. – Lepton flavor violations in τ decays. • Physics models can be identified (if new effects are observed) or new ones can be constrained (if nothing is seen). • Even in the worst case scenario (e.g. for MFV), B −> τν , D τν can probe the charged Higgs in the large tan β region. • Physics motivation is independent of LHC . – If LHC finds NP, precision flavour physics is compulsory. – If LHC finds no NP, high statistics B/ τ decays would be a unique way to search for the TeV scale physics. Marko Bračko DISCRETE`08, Valencia 5 M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 5

Introduction: KEK Site Introduction: KEK Site Tsukuba e + Tokyo e – Narita Mt. Fuji M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 6

Introduction: Introduction: Belle II Experiment Belle II Experiment ~600 members almost 100 institutions from 23 countries/regions M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 7

Introduction: Introduction: Belle II Collaboration Belle II Collaboration ~600 members almost 100 institutions from 23 countries/regions M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 8

Projection of Luminosity at SuperKEKB Projection of Luminosity at SuperKEKB Integrated luminosity Integrated luminosity Will reach 50 ab -1 in 2022 50x Belle ) -1 ) (ab -1 (ab Physics run starts 2016 40x Peak luminosity Peak luminosity Belle Commissioning ) -1 ) s -1 starts early 2015 -2 s Shutdown (cm -2 (cm for upgrade Year Year M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 9

Resource Estimates Resource Estimates → Similar data rate as LHC experiments (and comparable in total requirements)! M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 10

Belle II Computing Model Belle II Computing Model M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 1 1

Belle II Distributed Computing System Belle II Distributed Computing System ● Based on existing, well-proven solutions plus extensions for Belle II ➔ DIRAC for job management ➔ AMGA for metadata ➔ CVMFS for software distribution (thanks to CERN and Steve Traylen for providing the Stratum-0 server, and to GridKa for the stratum-1 server) M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 12

Workflow Abstraction Workflow Abstraction Output data files Input data files (analysis) jobs FileA FileX Job1 FileB FileY Job2 FileC FileZ ... ... ... Project Output dataset Input dataset ➢ Usage of different middlewares is treated by Dirac in a user-transparent way. ➢ Don't deal with single files and jobs, but with datasets and projects M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 13

Projects Projects ● Job submission  gbasf2 -r 1000 -s B2Kpi.py -p B2Kpi_s01 ● Job monitoring  gb2_project_summary  gb2_project_analysis --Project B2Kpi_s01  gb2_job_status --Project B2Kpi_s01 --Status=failed ● Rescheduling of failed jobs  gb2_job_reschedule --Project B2Kpi_s01 ● Job output  gb2_job_output --Project B2Kpi_s01 --Status=failed M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 14

MC Production Campaigns MC Production Campaigns ● 1 st MC Campaign: February 28 – March 19, 2013 ● 1st stage: event generation and detector simulation → raw data ● 2nd stage: reconstruction ➔ 240k jobs, 40 kHS*days ➔ 60M events, 190 TB of output data ~20% failure rate:  metadata registration, input data download, application errors M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 15

MC Production Campaigns MC Production Campaigns ● 2 nd MC Campaign: July 23 – September 8, 2013 ● Simulation and reconstruction, with background mixing → mdst data ➔ 630k jobs, 700 kHS*days First Second ➔ 560M events, Campaign Campaign 8.5 TB of output data Output data ~10% → 1% failure rate:  upload recovery site configuration/ downtime, proxy expiration, server load, human errors No crash of offline software  M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 16

MC Production Campaigns: MC Production Campaigns: KEK Contributing Sites (2013) Contributing Sites (2013) GridKa CESNET DESY Nagoya KISTI SIGNET UA-ISMA CESNET CYFRONET FNAL Virginia Nebraska PNNL M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 17

MC Production Campaigns: MC Production Campaigns: Contributing Sites Contributing Sites M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 18

MC Production Campaigns: MC Production Campaigns: Issues & Solutions Issues & Solutions DIRAC load ● Failover mechanism for output storage (Increased number of pool accounts on KEK SE) ● Proxy lifetime extended to 168 hours ● Communication frequency with DIRAC decreased and services distributed over more nodes M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 19

MC Production Campaigns MC Production Campaigns ● 3 rd MC Campaign: ~ April 1 – May 15, 2014 ● Simulation and reconstruction, with background mixing → mdst data ● 2x previous CPU#: 11k concurrent jobs; > 80 kHS max ● ~30 sites contributing ● 4.2G events produced → Very successful; also updated analysis and grid software → To obtain useful data for physics studies new extensive MC production started this week M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 20

Data Challenge ➢ Network connection between sites is essential  Raw data from KEK to PNNL  Mdst data and MC between sites world wide ➔ Transfer tests between different sites in May 2013 with FTS2 server PNNL at GridKa 21 M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 21

Monitoring 22 M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 22

Relation to the EGI infrastructure Relation to the EGI infrastructure ● Significant share of our resources is part of EGI infrastructure (current estimates: almost 70 %) ● Resources are provided by Resource Providers: sites from (HEP) institutions participating in Belle II, but also from non-membering institutions ● We have special contacts with GEANT to enable the necessary connectivity (relation to LHCONE has already been brought up in the discussion) ● Usage of central infrastructure: GGUS, GStat2, and GOCDB. ● So far, EGI infrastructure has proven to be essential for the Belle II computing model . M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 23

EGI: Future (additional) collaboration EGI: Future (additional) collaboration ● EGI contact network: Our development could benefit from contacts throught the EGI contact network, meetings. ● EGI facilities: → Future Belle II computing meeting(s) in Amsterdam? ● EGI support might be crucial in various funding proposals: → There is already a VLDATA DIRAC-based project proposal in preparation with other partners (EGI has played an important role in identifying possible partners and hosting application preparation events) ● EGI support for DIRAC is vital for us: We are relying on the long-term collaboration with DIRAC through which we should be compatible with future technical development of EGI (e.g. support for middleware deployed by DIRAC). M. Bracko: Belle II computing model EGI-CF , Helsinki, 2014-05-22 24