mcm monte carlo management service
play

McM Monte-Carlo Management Service Jean-Roch Vlimant for PdmV ( * ) - PowerPoint PPT Presentation

McM Monte-Carlo Management Service Jean-Roch Vlimant for PdmV ( * ) & Generator Groups CMSDAS @ FNAL, January 2014 https://twiki.cern.ch/twiki/bin/view/CMS/SWGuideCMSDataAnalysisSchool2014GeneratorExerciseatFNAL


  1. McM Monte-Carlo Management Service Jean-Roch Vlimant for PdmV ( * ) & Generator Groups CMSDAS @ FNAL, January 2014 https://twiki.cern.ch/twiki/bin/view/CMS/SWGuideCMSDataAnalysisSchool2014GeneratorExerciseatFNAL https://twiki.cern.ch/twiki/bin/view/CMS/PdmVMcM https://cms-pdmv.cern.ch/mcm/ (production) https://cms-pdmv-dev.cern.ch/mcm/ (test instance) ( * ) PdmV : Physics Data-MC Validation

  2. Global Picture Generator Contacts Production Manager Computing Operation  Operate McM  Prepare requests  Guide preparation of  Receive well formatted  Report special needs requests requests  Keep an eye on production  Dispatch production to sites McM  Operate at various steps  Follow through with sites  Provide validation issues  Provide book-keeping  Manage central space  Guide production 2 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  3. The Path To An Analysis Dataset ● Figure out the relevant generator parameters ● Possible modifications in simulation itself ➢ for better data/MC agreement ● Figure out the requirements in terms of digitization and reconstruction to match the analyzed data ● All steps can be done in the same workflow, however ➢ Computing requirements  Generation and simulation mostly done at T2s  Digitization and reconstruction mostly done at T1s ➢ Chronology of production  Generation and simulation may start before digi-reco  Several digi-reco might be required for a given sample of generated events ➔ Steps are split in several processing workflows ➔ Each group of steps is organized in a campaign 3 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  4. The Concept of Chained Campaigns 4 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  5. Terminology ● Campaign Campaign is a group of requests that share the same of very similar configuration, energy, software release series, physics performance … ✔ Example : Summer12 is for all gen-sim at 8TeV using the 5.3 release serie ● Request Request is a specific member of a campaign for the production of a given dataset, being intermediate or for analysis (AOD) ✔ Example : TOP-Summer12-00177 is the request for production gen-sim dataset of top pair production and decay in leptons, with top mass hypothesis of 173.5 GeV ● Flow Flow is a connector between campaigns, it defines specific operation that needs to be done while using the output of one request as input to its next step ✔ Example : flowS12to53RD defines the necessary ingredients to have a “run dependent” digi-reco of Summer12 gen-sim samples within Summer12DR53X campaign ● Chained Campaign Chained Campaign is constructed from campaigns that are connected together with flows ✔ Example : chain_Summer12pLHE_flowLHE2S12_flowS12to53RD (aliased to pLHE12DR53RD) is the chain of steps required to produce a run-dependent analysis dataset from a privately produced set of .lhe files ● Chained Requests Chained Requests is a specific member of a chained campaign, made from existing requests. ✔ Example : HIG-chain_Summer12pLHE_flowLHE2S12_flowS12to53RD-00001 is the chain of request required to produced the analysis dataset for a Higgs+top production with decay in a pair of photons 5 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  6. Terminology ● Campaign Campaign ● Request Request ● Flow Flow ● Chained Campaign Chained Campaign ● Chained Requests Chained Requests 6 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  7. Why All This ● Flows and chained campaigns are overkills if there is only one way of producing samples ✔ It used to be the case ✗ It is far from being true anymore (RD digi-reco, several pile-up, …) ● In the context of knowing what should happen for the production of a given dataset, flows, chained campaigns and chained requests allow ✔ The production manager to know what is next the next step to be done ✔ The generator contact to keep an eye on the evolution of the production through the different steps ✔ The user to have at a glance the history for the production of an analysis sample ● Not only McM is a production management tool, it also allows for book-keeping and documentation ✔ Generator parameters can be updated at anytime ✔ Notes should contain details of the content of the sample if there are some specificity ✔ The configuration files used for production is accessible ✔ The “cmsDriver” command is accessible ● McM offers extra protection to computing operation and the user themselves ✔ Runtest of the request ensures run-ability ✔ Validation step allows to scrutinize the generator content of a request 7 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  8. Evolution of a Chain of Requests 8 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  9. Practical Exercise : Request 1/2 ● Required ingredients  dsn : https://twiki.cern.ch/twiki/bin/view/CMS/ProductionDataSetNames ➔ Necessary to label the output datasets /<dsn>/<some string>/<data tier>  Time/event ➔ Mandatory to be accurate so as to not exceed job time-out in production  Size/event ➔ Required to be accurate to allocate enough space for the output dataset  Efficiencies ➔ Mandatory to be accurate so that the exact number of events get produced  Extension number ➔ Mandatory to guide the user in what need to be combined, and prevent dataset over-writting in production  Process string ➔ Mandatory when the request is slightly changed from the campaign by the user, and need to be properly label /<dsn>/<some string containing the request process string>/<data tier> ● Insert a request in McM ➔ Create a new request and insert the gen-fragment from previous exercise ➔ Select to be provided with validation plots ➔ Then run the local test : to verify that everything is in order ● Verify requests values ➔ Edit with correct values : to have all ingredients accurately inserted ● Toggle “validation” (and move on to next step) : triggers the runtest to be performed under McM 9 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  10. Practical Exercise : Request 2/2 ● Prepare a request for processing the request towards an analysis dataset ➔ Insert an MccM document for the request : simplifies the report at MC coordination meeting with exact ingredients ➔ Do not select the block # until really sure of the rest : to prevent the chains to be generated automatically as part of the exercise ● When validation is finished (watch for notification and change of status) ➔ Inspect validation (requires cmsweb authentication) : verify that the generator fragment is creating what is expected ➔ Toggle “define” : status defined means that it's read and validated ● Watch the requests getting automatically processed as part of the exercise ✔ Approved : the generator conveners has looked at the request and things seem to be in order ✔ Chained : the production manager has acted on the MccM document to create the required chains, and reserved the necessary requests ✔ Submitted : the production manager manually submitted the request or McM submitted automatically once chained and approved ✔ Done : after processing by comp-ops, the output dataset is set to VALID and the request is “done”. At warp speed for the purpose of the exercise, there is no real processing 10 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  11. Practical Exercise : Find Information ● Use the production instance https://cms-pdmv.cern.ch/mcm/ ✔ Find the status of certain requests ➔ /TTJets_MSDecays_matchingdown_TuneZ2star_8TeV-madgraph- tauola/Summer12_DR53X-PU_S10_START53_V19-v2/AODSIM ➔ /TTJets_MSDecays_mass178_5_TuneZ2star_8TeV-madgraph- tauola/Summer12_DR53X-PU_S10_START53_V19-v1/AODSIM ✔ Find the Xsec of a given dataset ➔ /TTJets_SemiLeptDecays_8TeV-sherpa/Summer12_DR53X- PU_S10_START53_V19-v1/AODSIM ✔ Find the generator fragment used to produce a given dataset ➔ /TprimeTprimeToTHBW_HToGammaGamma_M-900_TuneZ2star_8TeV- madgraph/Summer12_DR53X-PU_S10_START53_V19-v1/AODSIM ✔ Find the filter efficiency of the physics process in a give dataset ➔ /Xibstar0ToXibPi_8TeV-pythia6-evtgen/Summer12_DR53X- PU_S10_START53_V19-v1/AODSIM ✔ Find the “cmsDriver” command used to produce a give dataset ➔ /TTToHcWb_HToGG_8TeV-madgraph-pythia6/Summer12_DR53X- PU_RD1_START53_V7N-v2/AODSIM ● Find the same info through DAS ➔ All but config file and cmsDriver can easily be obtained by having DAS show the information stored in McM 11 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  12. Documentation Monte Carlo Coordination Meeting https://twiki.cern.ch/twiki/bin/view/CMS/PdmVMccM Main Twiki https://twiki.cern.ch/twiki/bin/view/CMS/PdmVMcM Presentation and Tutorials https://twiki.cern.ch/twiki/bin/view/CMS/PdmVMcMTutorials 12 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

  13. Summary of Exercise ● McM allows for submission of request to central production under central prioritization ● McM provides means of validation of the request content ● McM keeps track of the various required step of production ● McM provides book-keeping and documentation of the content of the analysis datasets ● McM information is aggregated and accessible through DAS 13 McM @ CMSDAS, Jean-Roch Vlimant 1/5/14

Recommend


More recommend