Monday January 17th Room 317, 3:30pm MDT

Attendees: Jeff, Sanjay, James, Rob, Brian, Kumar, Jim, Urvashi

Targets/tasks for January 10th - January 17th

• One off meeting 10am Tuesday January 11th to discuss async I/O approach
• Continue code development of async I/O, attempt to integrate with Clean Jim
• Time domain partitioning with multiple MSes per imaging process using semi-blessed test scripts Kumar
• Examine small I/O causes in Split Rob Reid/James/Kumar

Discussion/Agenda

• Continue code development of async I/O Jim
  • working toward getting imager to make use of it, difficulties in integration
  • make aipsrc variable to control async vs sync I/O
  • Discussion of visiter vs visbuffer
  • Proposal to have a one off meeting to discuss I/O access pattern 10am tuesday
• Complete remaining timing scripts Greg at AAS
  • Small spectral status (needed further cleaning)
    • does not include continuum subtraction as a separate timed step using uvconsub2.
  • Large continuum essentially done
  • Large spectral on hold
• Generate procedures and first timings for test cases James
  • Split seems much better
  • Tile size substantially affects performance, need to characterize behavior
  • Still need other data sets
• Time domain partitioning with multiple MSes per imaging process Kumar
  • Mention last week that cube case had regressed, is that real ?
  • Trying to identify 10% small reads, possibly metadata column (either main or sub table)
  • Send Kumar existing versions of test scripts for parallelization James
• Issues with image size and multiple threads. Urvashi
  • Some performance variance by image size
• Rob Reid in NM
  • Looking at split I/O characteristics
  • Recently changed to visiter based access instead of row by row access
  • Need to understand why Split and by extension other tasks make so many small I/O calls
  • Split also makes a large number of larger I/O calls.
• Cluster scheduler scripting with CV James on hold for scripts
  • Discuss with Jim semantics for cluster/CASA interaction
• Cluster resource definitions James/Jim on hold