1 / 18

New Seaborg Queue Configuration Results May 29, 2003 David Turner NERSC User Services Group

New Seaborg Queue Configuration Results May 29, 2003 David Turner NERSC User Services Group dpturner@lbl.gov 510-486-4027. Introduction. Review of LoadLeveler Class Structure NERSC-3 classes Proposed NERSC-3 Extended classes Current NERSC-3 Extended classes

alissa
Download Presentation

New Seaborg Queue Configuration Results May 29, 2003 David Turner NERSC User Services Group

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. New Seaborg Queue Configuration Results May 29, 2003 David Turner NERSC User Services Group dpturner@lbl.gov 510-486-4027

  2. Introduction • Review of LoadLeveler Class Structure • NERSC-3 classes • Proposed NERSC-3 Extended classes • Current NERSC-3 Extended classes • Objectives of current class structure • Effects of Current Structure • Connect time wallclock * nodes * 16 • Wait time start time - submit time • Connect time / Wait time • Conclusions

  3. NERSC-3 Class Structure

  4. Proposed Class Structure

  5. Other Proposed Changes • Various limit adjustments • Increase user run limit from 4 to 6 • Eliminate class limit of 7 in regular_long • Retain 1 running, 1 queued limit in regular_long • Eliminate aging • Incompatible with class priorities • Schedule lowest load average and smallest memory nodes first • Tune scheduling parameters to maintain responsiveness

  6. Current Class Structure

  7. General Batch Policies • Each user may have: • 6 jobs running • 10 jobs considered for scheduling (idle state) • 30 jobs submitted • The class run limit for reg_1l is 15 jobs • Jobs requesting 8 hours or less will complete before scheduled outages • Jobs placed on “user hold” (status HU) will be removed after one week

  8. Objectives of Class Structure • Allow 4096-way jobs • Current MPI maximum • Favor “large” jobs • Provide longer time limit for “regular” jobs • Provide more resources to “long” jobs • Allow greater access • Provide more resources to interactive and debug jobs • As needed All while maintaining system responsiveness

  9. N3 vs. N3E • N3 • October 1, 2002 – March 2, 2003 • 153 days • N3E • March 3, 2003 – May 20, 2003 • 79 days

  10. Jobs Per Week

  11. Connect Time vs. Class Charge Class

  12. Connect Time vs. Size Number of Nodes

  13. Wait Time vs. Size Number of Nodes

  14. Connect Time / Wait Time Number of Nodes

  15. Conclusions • Users running larger jobs • Users running longer jobs • Interactive and debug throughput maintained

  16. Resources I http://hpcf.nersc.gov/running_jobs/ibm/llsum/summary.php

  17. Resources II http://hpcf.nersc.gov/running_jobs/ibm/llsum/

  18. End of Talk This slide intentionally left blank.

More Related