Casa Parallelization Meeting Minutes

Thursday August 21st, DSOC 317, 8:00AM MT

• Polycom Video: 192.33.117.12##8108
• Voice: 434-817-6523

Attendees:

• Socorro: Rob, Jeff, Lindsey, James, Kumar
• Charlottesville:
• Garching: Justo, Sandra

• Apologies:

Discussion

• Review of Action Items from last meeting (see table below).

• Preferred MPI library for development environment.
  • MPI Implementation
  • OpenMPI, MPICH and MVAPICH use within NRAO. Limited ESO use of MPI.
    • Rob to ask Kana about experience with MPI
  • Hydra has broader support for resource manager integration (PBS, Torque, etc), but we may have multiple launcher configurations for different applications so this isn't a deciding factor.
  • Difference between MPICH and MVAPICH on infiniband support was tunnel tcp/ip vs native infiniband vs infiniband w/ RDMA. MVAPICH uses the MPI I/O libraries for speed. Realistically, we don't need this level of optimization.
  • Conclusion - after reviewing the requirements it appears that all the major implementations will meet our core needs. ACTION: Justo will test integration of openmpi on the AOC lustre cluster and review the listed requirements in detail. Assuming there are no identified deficiencies with openMPI, we will proceed with the mpi4casa as implemented.

• Review Task List for current cycle.
  • CasaParallelTaskList

• Status Updates
  • Current Status: Summary Page
  • Revisit inclusion of cvel2 in 4.3 release:
    • It may be possible to have a first version of cvel2 in 4.3.
    • Reassess inclusion/deadline as we approach the feature freeze date (9/15).

• Anything else you would like to discuss (AOB)
  • Justo and Sandra to visit NM this fall
    • Work with James, Kumar, B&T hire, etc, on 1st release issues.
    • Sandra noted that trip would have to be in the November/December time-frame.
    • November 3-Nov 7 Pipeline Parallelization visit to Edinburgh, UK.
    • Perhaps just extend Edinburgh visit and have a second working meeting there? (TBD)

• MSTransform heuristics when processing an input MMS
  • ACTION: Need to ask Jeff about 3rd case below - is it a reasonable use case?
  • This is how I am currently implementing it:

            Case 1: combinespws=True or nspw>1 and separationaxis != 'scan'
                  if spws are contained in all subMSs
                      process MMS in parallel
                  else
                      monolithic processing of MMS
                      set output axis = scan
                
            Case 2: timeaverage=true and timespan='scan' and separationaxis != 'spw'
                  if scans are contained in all subMSs
                      process MMS in parallel
                  else
                      monolithic processing of MMS
                      set output axis = spw
                
            Case 3: (combinespws=true or nspw>1) and timeaverage=true and timespan='scan' and separationaxis = 'auto'
                      if spws and scans are contained in all subMSs
                          process MMS in parallel
                      else
                          raise exception (because it cannot set numsubms for output)   
                      if separationaxis != auto
                          raise exception (because it cannot set numsubms for output)                                        

Deferred Items for a future meeting

• OSX support for MPI and OSX testing of MPI framework.

• Logger
  • Single log file & concerns regarding ordering of entries.
• Imaging
  • Discuss requirement imposed on VI/VB2 to pass required information for MMS processing.
• MSTransform
  • Discuss time separation axis for MMS creation.

Action Item List

Closed Item Record