-- SandraCastro - 2015-02-03

Casa HPC/Parallelization meeting agenda/minutes


Thursday [05/02/2015], [ESO Centaurus, C.2.01], [9:00 AM MST]


How to connect

  • Dial in: +49 89 307 6834
  • Video connection: 46014@eso.org (46014@134.171.42.27)

Attendees:

ESO: Julian, Justo, Sandra

Socorro: Jeff, Lindsey, James, Sanjay, Joe, Tak

CV: Andy, Mark

Agenda

  1. Build team: new test tarfile with MPI libs
  2. Pipeline testing with MMS in Garching (Sandra) and in Socorro (James)
  3. Lazy mode of importasdm with MMS. What's next to test?
  4. Balanced mode of partition
  5. setjy parallelisation

Minutes

1. Build

  • Lindsey reported the pipeline runs slower with the new casa binaries. Julian says all the new changes would not affect performance. The regression were all green with this new binary.

2. Pipeline testing

  • James is running the pipeline with MMS and the latest binaries in lustre. He reports that partition behaves as expected with no changes in I/O behavior compared to old partition.
  • Sandra reported that she run the pipeline which seg faults in clean with a backtrace to setTileCache() in the VisibilityIteratorReadImpl class.
  • Sanjay said clean would see an MMS as a monolithic MS and should work normally.
  • Sandra will try clean by hand on this MMS.
  • Sanjay saw an error in casacore when using the mpi compilers.
  • Sanjay He would like to push for using tclean and if possible, integrated with MPI. Justo will work on the parallel_go wrappers in the MPI framework next week.

3. Testing lazy mode and MMS

  • Justo reported that lazy mode and MMS work without any problems on a cycle2 script. He will create a small regression to compare every step with and without lazy option. The lazy import is much faster than the normal import.
  • Jeff said the pipeline will have a switch “parallel” to include importasdm with lazy option + partition.
  • Sandra said that it is straighforward to call ParallelDataHelper inside importasdm to create an MMS. We will look into this together with Michel in the near future. Things to look at when doing this: bdfflags, wvr_corrected_data MSs option.
  • Lindsey said the option to re-order the correlations in importasdm needs to be tested in combination with partition.

4. Balanced mode of partition

  • Sandra informed that there are still a few things to fix in the new function, mainly to support EVLA MSs that have repeated SPW IDs. As soon as this is fixed, the new mode will become the default in partition and mstransform.

5. AOB

  • Sandra informed that Julian has added all CASA’s functional tests in the Jenkins systems of the Pipeline System group at ESO. They run once a day. He also added a few tests running with MPI. We use these tests to react immediately if something breaks due to our changes and it is not a replacement of the CASA test system in CV. The Jenkins server here cannot be accessed from outside ESO.

Action Item List

  • Julian will check our local tests and see if there are signs of slowness with the latest binaries.
  • Andy will check regression run times for any performance differences with latest binaries
  • Sandra to ask build/test team to include test_mpi4casa in the list of smoke tests. It should also be included in the nightly tests. Sandra will send also a list of the tests that should run with mpirun in front.
  • JK check if other people in the team also prefer separated parallelisation logs

Topic revision: r6 - 2015-02-11, SandraCastro
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