250 likes | 373 Views
Software Testing. Michael P. Rupen EVLA Project Scientist for Software. Connecting Scientists and Software. Software is both complex and important defines telescope (e.g., rms noise achievable)
E N D
Software Testing Michael P. Rupen EVLA Project Scientist for Software EVLA Advisory Committee Meeting May 8-9, 2006
Connecting Scientists and Software • Software is both complex and important • defines telescope (e.g., rms noise achievable) • programmers are professionals – cannot assume astronomical knowledge or p.o.v. (e.g., declination -00:00:02) • still learning (e.g., user interface; implications of dynamic range goals) EVLA Advisory Committee Meeting May 8-9, 2006
Connecting Scientists and Software • organizational split: much higher walls between scientist and implementation • sheer size of project • accounting requirements • history of aips++ • requires close connections between astronomers and programmers • at all levels, from overall priorities to tests of well-defined sections of code EVLA Advisory Committee Meeting May 8-9, 2006
Connecting Scientists and Software:EVLA Approach • EVLA ProjSci for SW: Rupen • responsible for requirements, timescales, project book, inter-project interactions, … • EVLA SW Science Reqts. Committee • Rupen, Butler, Chandler, Clark, McKinnon, Myers, Owen, Perley, (Brogan, Fomalont, Hibbard) • consultation for scientific requirements • source group for more targeted work EVLA Advisory Committee Meeting May 8-9, 2006
Connecting Scientists and Software:EVLA Approach • Subsystem Scientists • scientific guidance for individual subsystems • day-to-day contacts for programmers • interpret scientific requirements for programmers • oversee (and are heavily involved in) tests • heavily involved in user documentation • consult with other scientific staff • duties vary with subsystem development EVLA Advisory Committee Meeting May 8-9, 2006
Connecting Scientists and Software:EVLA Approach • Subsystem Scientists (cont’d) • Currently: • Proposal Submission Tool: Wrobel • ObsPrep: Chandler (“daughter of JOBSERVE”) • WIDAR: Rupen/Romney • Post-processing: Rupen/Owen • TBD: Scheduler, ObsStatus Screens, Archive EVLA Advisory Committee Meeting May 8-9, 2006
Connecting Scientists and Software:EVLA Approach • Less formal, more direct contacts • e.g., Executor (e.g., ref. ptg. priorities and testing) • e.g., Greisen developing plotting programs based on Perley hardware testing • Testers • On-going tests: small in-house group (fast turn-around, very focused) • Pre-release tests: larger group of staff across sites and projects • External tests: staff + outside users EVLA Advisory Committee Meeting May 8-9, 2006
Science Domain Tests • Proposal Submission • ObsPrep • Archive • ObsMon EVLA Advisory Committee Meeting May 8-9, 2006
Science Domain Tests:Prop. Submission Tool • Testing timetable for this release: • Feb 10: post-mortem based on user comments • Mar 1: internal test-I (van Moorsel, Butler, Rupen) • Mar 20: internal test-II (+Frail) • Apr 18: NRAO-wide test (14, incl. CV/GB) • May 3: internal test-III • May 7: public release EVLA Advisory Committee Meeting May 8-9, 2006
Science Domain Tests:Prop. Submission Tool • rest of May: fix up & test post-submission tools • Early June: gather user comments • June 7: post-mortem, set next goals EVLA Advisory Committee Meeting May 8-9, 2006
Post-Processing(CASA) • Pipelines • Interactive processing EVLA Advisory Committee Meeting May 8-9, 2006
Post-Processing:On-going Interactions • NAUG/NAWG • each meets once a month • track and test current progress & algorithmic developments • Scistaff members: Myers, Brisken,Brogan, Chandler, Fomalont, Greisen, Hibbard, Owen, Rupen, Shepherd, Whysong EVLA Advisory Committee Meeting May 8-9, 2006
Post-Processing:On-going Interactions • ALMA formal tests • ALMA testers themselves often have VLA experience (e.g., Testi) • ALMA2006.01-4 included Brisken, Owen, Rupen; first look at python • EVLA goals: gain experience; learn how ALMA tests work; influence interface EVLA Advisory Committee Meeting May 8-9, 2006
Post-Processing:On-going Interactions • User Interface Working Group (Apr 06) • goal: refine requirements for user interface • Myers, Brogan, Brisken, Greisen, Hibbard, Owen, Rupen, Shepherd, Whysong • draft results on Web EVLA Advisory Committee Meeting May 8-9, 2006
EVLA CASA Tests:Philosophy • Planned opportunities for collaboration • testing for mutual benefit, not an opportunity to fail • important to repair aips++ programmer/scientist rift • allows us to reserve time of very useful, very busy people (Brisken, Owen, …) • Set priorities • Check current approaches and progress EVLA Advisory Committee Meeting May 8-9, 2006
EVLA CASA Tests:Philosophy • Provide real deadlines for moving software over to the users • Prepare for scientific and hardware deadlines (e.g., WIDAR) • joint decisions on topics • feedback going both ways (e.g., what requirements are most useful [cf. DR]) EVLA Advisory Committee Meeting May 8-9, 2006
EVLA CASA Tests:Philosophy • less formal than ALMA • avoid duplication • lesser resources • EVLA-specific work is often either research, or requires much programmer/scientist interaction (e.g., how to deal with RFI) • Learning curve for CASA is still steep. ALMA’s formal tests involve e.g. re-writing the cookbook; that is likely beyond our resources EVLA Advisory Committee Meeting May 8-9, 2006
EVLA CASA Tests:Current Status • May-July 2005 test: w-projection • first step in the Big Imaging Problem • generally good performance (speed, robustness) but lots of ease-of-use issues • Myers, Brisken, Butler, Fomalont, van Moorsel, Owen • Currently concentrating on • long lead-time items (e.g., high DR imaging) • H/W driven items (e.g., proto-type correlator) • shift to user-oriented system EVLA Advisory Committee Meeting May 8-9, 2006
EVLA CASA Tests:Current Status • Feb/Mar 2006 ALMA test • 1st with python interface • in-line help • March 2006 User Interface Working Group • tasks • documentation • scripting vs. “ordinary” use EVLA Advisory Committee Meeting May 8-9, 2006
EVLA CASA Tests:Fall 2006 • User interface • revised UI (tasks etc.) • revamped module organization • new documentation system • Reading and writing SDM • ASDM CASA MS UVFITS AIPS • Basic calibration, incl. calibration of weights EVLA Advisory Committee Meeting May 8-9, 2006
EVLA CASA Tests:Fall 2006 • Data examination • basic plots (mostly a survey of existing code) • first look at stand-alone viewer (qtview) • Imaging • widefield imaging (w term, multiple fields) • full pol’n imaging • pointing self-cal “for pundits” • Some of this may be NAUG-tested instead EVLA Advisory Committee Meeting May 8-9, 2006
EVLA CASA Tests:2007 • Test scheduled for early 2007 • probably focused on calibration and data examination • Driven by need to support initial basic modes (e.g., big spectral line cubes), and to learn from the new hardware (e.g., WIDAR + wide-band feeds) • Currently working on associated requirements EVLA Advisory Committee Meeting May 8-9, 2006
Involving Outside Users • Science domain • fast turn-around and active discussion make this a fair commitment • reluctance to show undocumented, unfinished product • current involvement is mostly at the end of the testing process EVLA Advisory Committee Meeting May 8-9, 2006
Involving Outside Users • CASA • focus has been on functionality, reliability, and speed • major advances, but documentation & UI have lagged • learning CASA for a test is a little painful, esp. since the UI is evolving rapidly EVLA Advisory Committee Meeting May 8-9, 2006
Involving Outside Users • algorithmic development is special: • many public documents & talks • interactions with other developers • student involvement: Urvashi; two from Baum • focus is shifting to user interactions • as this happens, we will involve more non-NRAO astronomers EVLA Advisory Committee Meeting May 8-9, 2006