Casa HPC/Parallelization meeting agenda/minutes


Thursday [6th Aug 2015], [ESO Centaurus, C.2.01], [15:00 UT]


How to connect

  • Dial in: +49 89 6834
  • Video connection: Use this one (46103@134.171.42.27).

Attendees:

ESO: Julian, Sandra

Socorro: Tak, Jeff, James

CV: Akeem, Andy, Mark

Agenda

  • Testing the interferometry pipeline(s)
    • Performance testing - James
    • Trunk testing, using different OMP_NUM_THREADS settings - Sandra

Breakdown per pipeline task (SERIAL run)

pipeline task OMP_NUM_THREADS = 1 OMP_NUM_THREADS = 2 %
hif_lowgainflag 0:12:34.625 0:15:13.066 17%
hif_setjy 0:13:53.944 0:21:33.646 35%
hifa_bandpass 0:07:39.390 0:12:12.070 37%
hifa_timegaincal 0:19:43.174 0:28:21.156 30%
hif_applycal 0:50:49.250 1:05:57.824 23%
hif_makeimages calibrator 1:56:27.090 1:24:14.663 27%
hif_makeimages target 5:05:16.875 3:29:39.605 31%
TOTAL time 21:37:37 21:08:38

Breakdown per pipeline task (PARALLEL run)

pipeline task OMP_NUM_THREADS = 1 OMP_NUM_THREADS = 2 %
hifa_importdata 0:26:04.045 0:23:48.286
hifa_flagdata 0:02:36.155 0:02:40.092
hifa_fluxcalflag 0:02:02.786 0:02:04.194
hifa_rawflagchans 0:05:57.273 0:06:09.129
hif_refant 0:00:30.048 0:00:30.435
hifa_tsyscal 0:00:51.783 0:00:46.698
hifa_tsysflag 0:02:03.757 0:02:14.165
hifa_wvrgcalflag 0:09:00.433 0:09:14.987
hif_lowgainflag 0:10:23.697 0:10:44.163
hif_setjy 0:10:25.833 0:10:01.369
hifa_bandpass 0:07:06.934 0:07:10.087  
hifa_timegaincal 0:18:17.221 0:19:01.781  
hif_applycal 0:27:16.840 0:33:10.634  
hif_makeimages calibrator 0:38:36.347 0:32:40.841  
hif_makeimages target 4:35:45.820 3:33:31.315  
TOTAL time 17:22:20 18:27:24  

  • Status of parallelisation of the pipelines for next release
  • Next meeting date
  • AOB

Minutes

  • Testing the interferometry pipeline - James
    • James reported that in his latest tests, tclean (calibrator imaging) is still slower in sequential when running on an MMS compared to a normal MS. He verified that tclean opens the SOURCE table 972 times for a MMS with 16 Sub-MSs. This test had OMP_NUM_THREADS = 1. This seems to be the same issue as with the FEED table.
    • James thinks there are other tasks with similar behaviour.
    • Sandra and Julian will verify this issue with strace.
  • Pipeline trunk testing with different OMP_NUM_THREADS
    • As described in the above tables.
    • Will re-run such tests on smaller MS and the pipeline.
  • Status of parallelisation of the pipelines for next release
    • There is no pending issue on the HPC development for cycle 4.5. With respect to tclean, Tier-0 is in place, but Tier-1 is still blocked by issues with writing the MODEL column back in parallel.
  • We will announce by email the date of the next HPC telecon.

Action Item List

  • Sandra will check if applying a .flagversions created from an MMS also works on a MS.
  • Julian should remove the parts of the MPI library that do not work properly in the CASA binaries (Example: mpicc, etc.)
  • Jim will look into the problem of the many reads of the FEED table.
  • Jeff will follow-up on the problem with too many fsync calls in CASA (CAS-7191)
  • James will run pipeline tests using different separation axis in partition (scan and spw), instead of 'auto' (balanced mode).

Topic revision: r4 - 2015-08-10, 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