1 / 14

Grid Collector

Grid Collector. Wei-Ming Zhang Kent State University John Wu , Alex Sim, Junmin Gu and Arie Shoshani Lawrence Berkeley National Lab In collaboration with Jerome Lauret, Victor Perevoztchikov, Valeri Faine, Jeff Porter, Sasha Vanyashin Brookhaven National Laboratory.

Télécharger la présentation

Grid Collector

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. Grid Collector Wei-Ming Zhang Kent State University John Wu, Alex Sim, Junmin Gu and Arie Shoshani Lawrence Berkeley National Lab In collaboration with Jerome Lauret, Victor Perevoztchikov, Valeri Faine, Jeff Porter, Sasha Vanyashin Brookhaven National Laboratory STAR Collaboration, July 2004

  2. A View of the Analysis Process • Users want to analyze some events of interest • Events are stored in millions of files • Files are distributed on many storage systems • To perform an analysis, a user needs to • Write the analysis code, run it • Specify the events of interest • Locate the files containing the events • Prepare disk space for the files • Transfer the files to the disks • Recover from any errors • Read the events of interest from files • Remove the files STAR Collaboration, July 2004

  3. Design Goals of Grid Collector Make analysts more productive by • Reading only events of interest • Automating the management of distributed files and disks STAR Collaboration, July 2004

  4. Approaches of Grid Collector • Allow users to specify events of interest using meaningful physical quantities • numberOfPrimaryTracks > 1000 AND vectorSumOfPt > 20 • Simplify step 2 • Automate file management tasks • Use File Catalogs to locate files • Use Storage Resource Manager to manage the disk space and file transfers • Remove steps 3 -- 8 STAR Collaboration, July 2004

  5. Query estimation / execution requests Query Estimator (QE) Bitmap index User’s Application open, read, close Query Monitor (QM) Caching Policy Module Disk Cache file purging file caching Cache Manager (CM) file caching request File Catalog (FC) Storage Access Coordination System Strength – • Allow user to specify events as range conditions • Automate file management tasks Weakness – • Designed for Objectivity data • Access only one HPSS STAR Collaboration, July 2004

  6. Grid Collector Coordinator Analysis code 2 3 1 Bitmap index In: conditions Out: logical files, object IDs File Locator In: logical name, Out: physical location File Catalog 1 New query 4 Event iterator 5 6 File Scheduler In: physical file File Catalog 2 11 Index Builder In: STAR tag file Out: bitmap index administrator Fetch tag file Load subset Rollback Commit 7 8 HRM 1 9 DRM NFS, local disk 10 HRM 2 Grid Collector: Architecture Clients Servers STAR Collaboration, July 2004

  7. GC Select events with range conditions Read only selected events Automate all file and space management tasks Scheduler Specify a list of files on disk Read all events of the files Use Data Carousel for HPSS files GC vs. STAR Scheduler Both can split large jobs to multiple machines STAR Collaboration, July 2004

  8. GC Use multiple File Catalogs and multiple Storage Systems Integrate index building functions into the server Improves index building speed Make use of distributed disk caches, clients can have their own caches STACS Limit to only one File Catalog and one Storage System Use a separate Index Feeder to digest tag files Has very low data transfer rate through CORBA Make use of one disk cache, clients must access the disk cache GC vs. STACS Both select events with range conditions Both automatically manage files and disks STAR Collaboration, July 2004

  9. This Year Process all files, including MuDST Build indices fast Use automated file management functions Indexing 15 million events took one week Interact with multiple File Catalogs Last Year Process event files, but not MuDST Build indices slowly Index feeder requires manual file transfer Indexing 5 million events took 10 weeks Interact with only one File Catalog This Year vs. Last Year STAR Collaboration, July 2004

  10. What Can Grid Collector Do For You • If you gather statistics on lots of events • Grid Collector allows you to work with files not already on disk • If you search for rare events, Grid Collector allows you to • Specify the events with ease • Access only relevant files • Read only selected events • If you want to try some analysis ideas outside of the main computer centers, • Grid Collect manages file and space for you STAR Collaboration, July 2004

  11. How To Use The Grid Collector • Must use StIOMaker • StIOMaker can now handle all files including MuDST • Replace StFile with StGridCollector • StIOMaker requires a StFileI object • One currently uses “new StFile(…)” to create a StFileI object • Grid Collector provides a new way, “StGridCollector::Create(SELECT geant, event WHERE …)” • Iterate through events as usual STAR Collaboration, July 2004

  12. How To Use -- More Details • External dependencies • Globus, ROOT, STAR Software • Storage Resource Manager (DRM, HRM) • ORBACUS • Servers • Main Grid Collector Coordinator • DRM/HRM • File Catalogs • Client library • User need to load this in the macros STAR Collaboration, July 2004

  13. How To Select Events • SELECT [MuDst|event|…] WHERE NV0>100 AND … • The WHERE clause consist of range conditions joined with logical operators AND, OR, NOT. • All tags and a few File Catalog key words can be used in the WHERE clause • Variables with multiple values can be addressed with index, e.g., scaAnalysisMatrix[7] STAR Collaboration, July 2004

  14. Status Of Grid Collector • One version in production mode at BNL • An updated version in final testing stages • Brave early adopters still needed • Contact information • Wei-Ming Zhang zhang@hpacq.kent.edu • Jerome Lauret lauret@bnl.gov • John Wu John.Wu@nersc.gov STAR Collaboration, July 2004

More Related