Casa Parallelization meeting minutes

Monday March 26th Room 317, 3:30pm MDT

Attendees: James, Jeff, Kumar, Sanjay,

Targets/tasks for March 19th to March 25th

  • Test model data column changes James

Discussion/Agenda

  • Initial cluster scaling calculations
    • Need to better understand effects of data size, image size, scales, terms and channels on MS-MFS and pcube
    • Discussed cube case serial gather step, time to lock using CASA locks is over 30 seconds, time to process channel is 50 seconds.
  • Casa 3.5 targets
    • Parallelize multi-field imaging
    • Parallelize MFS terms > 1 in the gridding step
    • Parallelize Cube imaging scatter gather
    • Expand threaded gridder case to >4 threads
    • Implement core affinity for threaded gridder case
    • Debate regarding benefits of asp vs moasic work (unrelated to each other but similar in labor)
  • Cluster object specification and its effect on tasks. How do tasks make use of the cluster object (node, memory and core counts).
    • Should enumerate max cores and memory per node
  • Memory requirements as a function of imaging algorithm and image size
  • MS-MFS memory handling (how does it handle over mem situations)
    • test memory and temp image use
  • Threaded gridder benchmark update James
    • Jeff to ask Scott to build active branch for testing
  • Update on Partitioning status (null selections, non-conforming tasks) Jeff
    • No update
  • Update on asdm2ms_xl schedule for stable inclusion
    • Still serial, eventually multi-threaded across spw, multi-process across scans
      • now considering multi-process across spw with shrdmem (single node only)
  • Model data tests
    • James to get with Kumar
  • Pcube and pclean conversion to task status Kumar
  • Revisit async iterator for writing Jim/Justo
    • Refactoring seems to be working in active
    • Justo is testing flagging appears to be an improvement
    • Attempt to make async I/O in flagging make it to release
  • Memory resident subtables Jim
    • Test performance James
    • Check that it affects number of FDs as well
  • Spectral imaging Kumar
    • Understand channel chunk choice effect on memory James
    • Obtaining data set from Drew to test
  • Need 400+GB continuum case with complex field
    • Stalled for partitioning

CASA HPC Initiatives for the 3.4 Cycle

  • Memory Footprint Issues
    • Deployment of Multi-Field MS-MFS (Urvashi) Deferred
    • Deployment of Parallel Gridder and splitting of the Grid (Jim)
    • Design of MS-MFS Minor Cycle Parallelization (Jim et. al.) using Friday meetings Deferred
  • Initial Deployment of ASP Clean (Sanjay)
  • Performance Issues:
    • Virtual Model Data Column (Kumar) Done
      • Removes model data column
      • removes need for Setjy
    • Memory Resident Subtables Deployment In progress needs testing
    • VisBuffer Processing Framework
      • Add Write capability (Jim) Need to revisit but seems to work in not imaging
      • Deployment in tasks (Jeff to Assign)
    • Parallel Filler (and partitioning) (Tak) In progress
  • User Interface:
    • Parallel Imager in Task (Kumar)
    • Parallelization of other Tasks (assigned to Europe)

Deferred Topics

ARDG Targets

  • What is negative side effect of CVel in filler, non-reversible
  • What needs to be done to acceptably implement Auto-boxing with large number of channels

-- JamesRobnett - 2012-04-09
Topic revision: r1 - 2012-04-09, JamesRobnett
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding NRAO Public Wiki? Send feedback