1 / 26

LHCb distributed analysis and ARDA

LHCb distributed analysis and ARDA. A.Tsaregorodtsev, CPPM, Marseille. ARDA Workshop, 21 January 2004, CERN. Outline. LHCb distributed analysis tasks Distributed analysis tools Analysis of the DC2004 data What we expect from ARDA Conclusions. Analysis tasks.

lula
Download Presentation

LHCb distributed analysis and ARDA

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. LHCb distributed analysis and ARDA A.Tsaregorodtsev, CPPM, Marseille ARDA Workshop, 21 January 2004, CERN ARDA Workshop, 21 January 2004, CERN

  2. Outline • LHCb distributed analysis tasks • Distributed analysis tools • Analysis of the DC2004 data • What we expect from ARDA • Conclusions ARDA Workshop, 21 January 2004, CERN

  3. Analysis tasks • LHCb distributed analysis tasks have (almost) no specific features compared to other LHC experiments: • As formulated in the HEPCAL II document; • In the following the batch analysis tasks are discussed mostly; the interactive analysis is currently limited to PAW/ROOT sessions. ARDA Workshop, 21 January 2004, CERN

  4. Analysis task flow Metadata catalog File catalog AlgOptions catalog Select Datasets Edit AlgorithmFlow Edit AlgParamOptions AlgFlowOptions DatasetOptions DLLs AlgParamOptions Job JDL InputData Job requirements OutputSandbox N-tuples Histos, logs Prepare Sandbox, JDL Submit Job InputSandbox DLLs JobOptions FileCatalog slice ARDA Workshop, 21 January 2004, CERN

  5. Analysis tools • GANGA User interface; • Metadata Catalog; • File Catalog; • Data Management; • Workload management; • Various information services: • Configuration service; • Job and System monitoring; • Software package manager; • etc ARDA Workshop, 21 January 2004, CERN

  6. GANGA User interface Job script JDL file Job Options file Submit job Grid/Batch System Gatekeeper Worker nodes Send job output Files Transfer Storage Element Get Monitoring Info Get job output Send Local Client Ganga Job object Ganga Job object Ganga Job object Ganga Job object Ganga Job object Job Factory (Job Registry Class) Job Options Editor Data Selection (Input/Output Files) Strategy Selection Job Requirements (LSF Resources, etc) Database of Standard Job Options Strategy Database (Splitting scripts) ARDA Workshop, 21 January 2004, CERN

  7. GANGA User Interface • GANGA will allow user to perform standard analysis tasks: • Selecting data making queries; • Configuring jobs, defining the job splitting/merging strategy; • Submitting jobs to the chosen grid resources; • Following the job progress; • Retrieving the job output; • Job bookkeeping. ARDA Workshop, 21 January 2004, CERN

  8. Services based architecture • The UI provides user with an access to various grid services: • Catalogs; • Job execution; • Information; • Monitoring; • These services are either developed in DIRAC – the LHCb distributed production system • Or imported into DIRAC from other projects • Or provided by LCG for tasks executed on the LCG2 platform. ARDA Workshop, 21 January 2004, CERN

  9. DIRAC services architecture Job Auditing Provenance Information Service Authentication Authorisation User Interface API Accounting Metadata Catalogue Grid Monitoring File Catalogue Workload Management Package Data DIRAC components Manager Management Other project components: AliEn, LCG, … Storage Computing Element Element Resources: LCG, LHCb production sites ARDA Workshop, 21 January 2004, CERN

  10. Bookkeeping database • Corresponds to the ARDA Job Provenance DB and MetaData Catalog; • ORACLE with XML-RPC service interface; • Very flexible basic structure with regularly built views to enhance standard queries efficiency. ARDA Workshop, 21 January 2004, CERN

  11. File Catalog • Used to be part of the Bookkeeping database, we never had the intention to develop our own FC; • Starting from DC2004 we are going to use AliEn File Catalog wrapped as web service with XML-RPC (or SOAP) interface; • Use ARDA File Catalog as it will become available; • Making POOL interfaced to the ARDA Catalog is necessary. • Using File Catalog slices as POOL XML catalogs to allow execution of applications without access to remote catalog. • For data produced in LCG we will (try to) use RLS: • Synchronization of the 2 catalogs is an issue. • Not all the MSS’s are ready to be used now ! ARDA Workshop, 21 January 2004, CERN

  12. Data management tools • Using bbftpd server as a SE: • Too limited the functionality; • Using passwords or globus certificates for authentication; • Considering other Data Management tools candidates: • AliEn Data Management; • LCG replica manager. ARDA Workshop, 21 January 2004, CERN

  13. Information services • Configuration service: • MySQL with an XML-RPC interface to provide DIRAC configuration parameters to other services and agents; • PackageManager service: • MySQL with an XML-RPC interface to provide information on software packages availability, dependencies and status; • Job Monitoring and Accounting; • Provides information via XML-RPC interface to job monitoring and accounting report generation applications. • System Monitoring service: • Monitoring the availability of the services and agents; • Looking at Instant Messaging (Jabber) mechanism of services presence detection. ARDA Workshop, 21 January 2004, CERN

  14. Software installation • In DIRAC we install the production software on the fly by the running job: • Proved to be very efficient and useful; • Only the first job of the kind installs the new version on a DIRAC site; • This worked as well on the EDG testbed: • Each job installing the software for itself; • On the LCG we will use special jobs and info tags to install the production software on the LCG sites: • Following the procedure proposed by the LCG team: • Software distribution uploaded to one of the SE on the grid; • Installation/verification jobs by a specially defined user account • User specific analysis algorithms are shipped as DLL’s in the InputSandbox of a job. ARDA Workshop, 21 January 2004, CERN

  15. DIRAC use of computing resources • DIRAC design goals to facilitate operation in various environments: • Scheduling jobs to “any grid” computing resources: • “native” sites, running DIRAC Agents; • EDG/LCG grid as a whole, passing through the RB; • EDG/LCG CE’s and SE’s as DIRAC resources. • Using “any grid” storage resources: • Data Management component to be able to replicate data between LCG and DIRAC storage elements; ARDA Workshop, 21 January 2004, CERN

  16. DIRAC WMS architecture ARDA Workshop, 21 January 2004, CERN

  17. DIRAC WMS components • Job Receiver: • Connects to the Job Data Base • Parses the JDL • Inserts the JDL into the appropriate tables • Notifies Optimizers that a new job has arrived via the Jabber Instant Messaging system • Optimizers • Extracts new job from DB • Inserts it into Job Queue with a particular rank • Sort Jobs • Matcher • Does match making between Job Jdl and resource jdl • Agent • Asks job to Matcher with its specific resource JDL • Send job to CE ARDA Workshop, 21 January 2004, CERN

  18. DIRAC WMS technologies • JDL Job description ; • Python as development language ; • Sharing developments with GANGA, maybe reuse some modules; • MySQL for Job DB and Job Queues ; • Condor Classad library for matchmaking (wrapped in python with SWIG) ; • Internal interfaces by instant messaging technologies ; • Python/Java Jabber ; • Advantages : Asynchronous message, distributed service, XML based, scalable, robustness, and so on… • Various batch system back-ends (CE’s): • LSF, PBS, Condor, EDG/LCG, fork. • Running jobs communicating with various services: • WN connectivity is mandatory ! (at least Outbound). ARDA Workshop, 21 January 2004, CERN

  19. OGSI compliant prototype • Development the DIRAC WMS interface as an OGSI compliant service: • Using GT3 toolkit • First experience: • The GT3 based services are difficult to develop, deploy and configure; • The service efficiency is low: • 1 minute per job submission instead of 1 second with the “standard” DIRAC service; • Looking for other OGSI implementations: • Lighter and easier to use, may be with somewhat limited fuctionality; • E.g. PyGridWare package. ARDA Workshop, 21 January 2004, CERN

  20. OGSI philosophy vs GT3 implementation Concepts Implementations OGSA – idea “Physiology of the Grid” Dirac Service GT3 – Globus OGSI implementation OGSI – formal standard (v1.0) Axis - Web Service framework Web Services – foundation for OGSI Tomcat – java application server ARDA Workshop, 21 January 2004, CERN

  21. Distributed analysis on the LCG • The analysis will use LCG as it is: • Task preparation in GANGA; • Submission to the RB of the LCG; • Using LCG SE/RLS/RM for data manipulation; • Making data produced outside LCG available for the LCG jobs is an issue: • Copying to LCG SE’s or just registering in the RLS ? • During the EDG tests, we studied the mode of operation when jobs were submitted directly to CE’s bypassing the RB • Hope this will be not necessary with more stable RB of the LCG; • Keeping this possibility in our toolkit. ARDA Workshop, 21 January 2004, CERN

  22. Batch versus Interactive analysis • All said so far concerns Batch Analysis: • Submitting jobs to batch system; • No direct interaction with the jobs. • We are interested though in the possibility to run interactive jobs on the grid: • Many issues to be solved: • Job efficient parallelization; • Resources reservation; • WN connectivity problems; • Etc, etc • The PROOF system is very promissing: • Allthough not applicable to LHCb straight away ARDA Workshop, 21 January 2004, CERN

  23. What we expect from ARDA • Lacking components, most notably: • File Catalog; • Data management; • Grid monitoring; • Authentication/authorization; • The architecture where we can choose the most suitable and best performing component implementations. ARDA Workshop, 21 January 2004, CERN

  24. What we expect from ARDA • More efficient development process: • Rapid development cycles; • Keeping the functional core and adding functionality incrementally; • Emphasis on intensive testing while the development. • Concurrent development of components to try out different ideas and to enhance the quality by competition. ARDA Workshop, 21 January 2004, CERN

  25. How we can contribute to ARDA • Participate to the definition of the services interfaces, testing and feedback; • Prototyping ARDA components using OGSI compliant implementations; • Developing the DIRAC WMS into an ARDA compliant service ARDA Workshop, 21 January 2004, CERN

  26. Conclusions • The first tests of the distributed analysis will be done during the DC2004 (May) using LHCb developed and LCG tools; • We see the further evolution of the LHCb distributed analysis tools within the context of the ARDA architecture and the proposed development process. ARDA Workshop, 21 January 2004, CERN

More Related