Monday Morning Meeting 20 Apr. 2015

  • DIAL-IN NUMBERS & PASSCODES:
  • USA Toll Free Number: 866-901-8266
  • USA Toll Number: +1-203-566-3863
  • PARTICIPANT PASSCODE: 725760

Attendance

  • Socorro: Jeff, Lindsey, Jim, Tak, Susan, George, Pam, Urvashi, Martin
  • CV: Mark, Darrell, Akeem, Andy, Dave, Ville
  • Garching: Julian, Sandra, Justo, Dirk
  • SCO:

News / Meetings / Visitors

Build, Release, Testing

  • 4.3.1 Problem understood but not yet corrected
  • 4.4 Still have an issue on the Linux Side (sdreduce)
    • wvrgcal issue on Mac (removed forward decln and namespace issues)
    • Vi/VB2 iterator possible fix is to clean up memory allocation.
    • plotms shows difference between tarball and developer build
  • Drew Brisbin working on plotms testing

Architecture

  • Technical Forums: Please read Jim's writeup on how we deal w/ casacore.

HPC

  • Minutes of meeting from last Thursday: https://safe.nrao.edu/wiki/bin/view/HPC/Apr1615
  • Solved the gaincal problem with MMS: It was related to bad performance of the OpenMP mods introduced in casacore, for pre-iteration sorting, that perform really bad with MMS in a loaded env.
  • Also noticed that OpenMP keeps consuming lots of CPU after a parallel region, so we again recommend to not use it even for sequential run, or at least reduce the iddle-CPU consumption after a parallel region. The true problem is that we have only one small parallel region in casacore, otherwise it would be ok. This does not apply to code which is fully parallleized like the imager.

Development

  • 4.5 Planning coming out

Pipeline

  • Development
    • calibration: spectral index handling heuristics
    • imaging: tclean mfs and frequency region selection
    • plotting: improvement to plotting classes for grouping receiver band, baseband, spw plots
  • CASA 4.3.1 / ALMA pipeline
    • preparation for final acceptance testing
    • simple spectral window mapping heuristics testing
  • CASA 4.3.1 / EVLA pipeline
    • ongoing acceptances testing, setjy differences between 4.2.2 and 4.3.1
  • HPC testing
    • open file issues, plotcal, code, setjy MMS problem
  • CASA 4.4 testing
    • plotms efficiency, general 2~3X slowdown, occasional pathological slowdown
  • quality of ALMA ASDM statistics

AOB

  • Build Tools: need the documentation to be in a central place.
    • Mark to create the "blessed repository" and remove out of date pages.

Developer Reports

Monday Meeting
  • Sanjay Bhatnagar
  • Sandra Castro
    • I have checked flagdata when running the ALMA pipeline, because Brian suspected it was leaving sub-tables open. I didn't find anything, but saw that there two files open in the cache at the end of the pipeline and they were probably left open by gaincal. Brian confirmed that flagdata was ok and found the problem to be on a call to a table query.
    • I am also verifying the fix from Pam on plotms, which was keeping a write-lock while running the ALMA pipeline.
  • Lindsey Davis
    • Started development of spectral index handling heuristics
    • Fixes to simple spectral mapping heuristics, discussion with PWG and George
    • Investigated linearPD interpolation option, discussions with Jeff and George
    • Investigated EVLA pipeline setjy issues with Brian K.
    • Investigated plotms efficiency issues with Pam and Jim
  • Pam Ford
    • CAS-7428 - ensured that MS files opened by plotms were closed at completion; under test by Sandra
    • continued plotms performance tests, inc. with tarball (tarball run took 2x developer release build)
  • Kumar Golap
    • Struggled to rebuiild casa on RHEL 6 for more than a day
    • talked to HPC group about multithreading gridders
    • Hunted a bug for Urvashi in setting weight density in ref- imager which ultimately lead to being a bug in defining the weight sent.
  • Justo Gonz├ílez Villalba
    • Fixed antenna reindexing problem reported for SPW separation
    • Setup MAC OS X 10.10 / Darwin 14 CASA development environment in the ESO MAC server
    • Fixed iPython-cluster security issues and system command backwards incompatibilities reported for MAC 10.10
    • Fixed split segfaults reported for MAC 10.10
    • Solved the gaincal problem with MMS: It was related to bad performance of the OpenMP mods introduced in casacore, for pre-iteration sorting, that perform really bad with MMS in a loaded env.
    • Discussions on new requirements for the parallelization framework to support external task parallelization.
  • Jim Jacobs
  • Jeff Kern
    • ICRAR Visit
  • Susan Loveland
    • CARTA
      • Selenium testing and tests of Commands
      • Save state work
    • CASA/Viewer/Profiler: CAS-7207
  • David Mehringer
    • Fixed imagetest failure
    • Discussions with Nathan re: channel widths
    • (CAS-7467) specifying spw by its name in tasks does not work: Punted to Mark as I'm not responsible for ms.select().
    • Was reminded by casa-staff thread that our build process is quite convoluted and fragile.
    • (CAS-7464) someone broke the spectrum of a component: Provided unsolicitied advice to ticket Reporter/Assignee to use cl.close() after calling cl.rename() in order to flush the contents of the component list to disk. Pointed out that this is documented in the inline help of cl.rename().
    • (CAS-7473) Fix raw pointer issue in LatticeApply and StatsTiledCollapser: This was an insidious defect that predates me in which the original implementation of LatticeStatistics implicitly assumed the arrays it was retrieving from a LatticeIterator were always contiguous. This was not always the case however. For example, the call to LatticeIterator::cursor() using a SubImage of a TempImage in which the region does not cover the entirety of the parent image will not produce a contiguous array. This defect has been silently producing incorrect statistics in these cases for years (and was the root cause of the imagetest regression failure earlier in the week). I modified the very old code to check for array contiguity, and if the array is not contiguous, to make a copy of the array (which will be contiguous) and use that for the computations.
    • Reinstated a large number of LatticeStatistics tests which had apparently been removed during the casacore alignment.
    • Commented on Jim's casacore process document, suggesting that there must be a process for deprecating and removing interfaces. Keeping poor and obsolete interfaces around simply for the sake backward compatibility in perpetuity is certainly not a good practice.
    • CAS-7459 Odd error when feathering. Failed AlwaysAssert: Gave unsolicited suggestion that feather check for unsupported use case and emit a user-friendly error message rather than producing an exception generated far down in the call stack that is meant for API developers and hence, undecipherable by most users.
    • (CAS-7463) msmd.chanwidths does not give error when unit is not a frequency unit: Fixed. As part of the fix, reinstated changes I made to Quantum in early February that were apparently lost during the casacore alignment.
  • George Moellenbrock
    • Found problems in plotms frequencies when averaging, submitted CAS-7460 w/ test data.
    • Some dev build maintenance related to libcasa_python and build config.
    • Diagnosed Q,U setting problem in setjy, wherein the sign of Q and U was always forced positive. Passed fix on to Tak, who is working on various other setjy issues.
    • Various polarization-related user-support, including TEC (Chris Hales), EVLA w/out parang coverage (Stan Kurtz), and ATCA linears via ALMA heuristics (Mark Weiringa).
    • Began retrieving data for uvcontsub failure: CAS-7478.
  • Dirk Petry
  • Martin Pokorny
  • Urvashi Rao
    • Just some user support emails and tests. Spent week on mostly non-casa work.
  • Darrell Schiebel
  • Ville Suoranta
  • Julian Taylor
    • tried to reproduce gaincal slowness, seems to depend on openmp settings and node oversubscription due to poor parallel quicksort in casacore
  • Takahiro Tsutsumi
    • Follow-up discussions (and how to test) with a tester on makemask bug fix (CAS-7302)
    • Looked into a couple of setjy issues
      • Investigated the difference in the pipeline results between 4.2.2 and 4.3.1. In the 4.3.1, the flux density used for scaling the model image seems be that of at the lower edge of the spw rather than the center frequency when the data is averaged to form a single channel data
      • The setjy error in MMS - seems to be some issues in defaulting of sub-parameter, still investigating
      • Checked in George's fix to bugs that he found in the polarization model setting (CAS-7434), he also noted usescratch=False is not working currently
Friday NAOJ Meeting
  • Kanako Sugimoto
    • tested fixed libsakura on RHEL 6 and OSX 10.10.
    • tsdbaseline: accept per spectrum baseline parameter ( CAS-7003): fixed sorting issue. developer tests showed good results. so bascally ready for user testing. I'm currently implementing unit tests.
    • Review of CASA 4.4 development cycle. Got feedbacks from developers.
  • Wataru Kawasaki
    • CAS-6999: the previous bug fix I made in libsakura was found not perfect; another problem was found when the left-most part of spectrum is flagged. fixed bugs by greatly simplifying the codes, and committed to Sakura's casa_44 branch again (r1351). Takeshi then updated libsakura tarball and sent it to Darrell.
  • Masaya Kuniyoshi
  • [CAS-6995] verified CORRECTED_DATA[i, j]=DATA[i,j] X FPARAM[i,j] .
  • [CAS-6995] verified WEIGHT is the median of WEIGHT_SPECTRUM on the channel axis.
  • Takeshi Nakazato
    • No developments
    • Reported an issue of OTF raster mapping. The start and the end of each raster row are flagged by online flag command with reason of Mount_is_off_source. It sometimes dropps edge data completely so that resulting map is effectively smaller than expected.

-- KanaSugimoto - 2015-04-16
Topic revision: r17 - 2015-04-20, JeffKern
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