Planning Meetings

September 26th, 2016 Debrief meeting

  • Attendees: Lois, Mauro, Vijay, Carol, Sherry, Brian
  • FYI...ATEPSC will be considered part of the ECP program in the future
    • May impact the agenda
  • Lois
    • ratings seem to show a pretty good balance
    • continue to struggle with hands-on
    • apparently needed more controversial panelst
    • did a little better on level of difficulty this time around
    • have a broad range of students and its hard to handle the mix
  • Mauro
    • was a new experience
  • Vijay
    • interesting that most thought the content was pitched at the ideal level
    • maybe take some lessons learned from how advanced it was
    • got several questions from students about meshing (two students stayed in touch via email)
    • Suggestion: Use survey to design a questionare to send to students 2-months prior to program next year
      • When students accept attendance to program, they have to fill out a survey about exptectations and then circulate this information to lecturers so they can adjust content to match expectations
  • Carol
    • good overall
    • looks like survey results are pretty good
    • do struggle with hands-on
    • SUNDIALS feedback hard to guage because of wide range audience
    • is there a chance that we cannot have the weekend?
      • Lois: It is hard to switch with other groups. They are all vying for an ideal spot
      • But, maybe best to make a case like Brian suggests about progression of topics
  • Sherry
    • reviewed feedback and have two thoughts
    • seems that students want more lecture-like material and some want more tutorial-like tutorial
    • What is the balance between tutorial-level content (e.g. how to use this package) vs. lecture content (e.g. theorey behind the package)
    • Need to work with ATPESC planners to assess these needs prior (see suggestion by Vijay, above)
    • In the past, putting in a lot of basic/fundamental stuff worked ok but that was removed this year to meet time-constraints
    • All of the hands-on are well-cooked examples. But, students want a "template" so they can do some more coding work
    • Have a standard problem or two that all packages can be applied to and let students see how each package is used to solve the same problem
      • This is similar to Jed Brown's idea from a couple years ago. He devied a 2D porous media flow problem.
  • Brian
    • crowd responded well to presentation (got a lot of eye contact); seemed engaged
    • might have been able to pitch more complicated material but half the crowd might not have been interested in that
    • thought the panels went over well
    • seemed grumpy about what versions of "parallel productivity" tools FASTMath is using (MP, MPI, etc.)
      • Fancier techniques and technologies should be put off to later in the 2-week session
    • Maybe FASTMath should come earlier in the 2-week session
      • Its kind of a buy or build issue
        • Here is what is available for "buy" the solution (early on in the training)
        • Here is what is availabe for "build" the solution (later in the training
  • Mark
    • Many survey comments regarding hands-on not being too useful because it was more of a compile-and-run rather than something that explains how to use the software
      • This is consistent with Barry's suggestion prior to this ATPESC for lecturers to focus more on how to use their packages than on what a package is capable of
      • Its possible this particular batch of students is more advanced than ever before and so hands-on exercies as they have worked in the past aren't as useful
    • Was surprised by comments regarding panels not being contorversial enough
      • I do think there was a diversity of opinion on the panels and that is a good thing
      • I don't think the panels need to look like "reality TV" to be useful or engaging

July 19th, 2016

  • Attendees: Lois, Carol, Vijay, Sherry, Cameron, Mark S., Rob F., Brian, Lori
  • Software install status
    • Mark to finish lib installs by end of week
    • Cameron finished with PUMI stuff
    • Glen/Mauro to install to Albanay
  • 3-Slide overviews
    • Just make sure people are moving ahead on these
    • Should be pretty quick and easy to just revise from slides used in previous years
    • Remember: these are all bundled together and presented and part of a long term record on line
  • Drafts of Presentations
    • In the past, we've had negative feedback about amount of material crammed into slides
      • Pls take care to really choose what issues to discuss in slides
      • In other words, less-is-more in the presentations
      • Also, remember the material is being recorded
    • Speakers should have final material available at the latest the day of the presentation
  • Panels and Panelists
    • A number of people are available for both days
    • Will divide folks up in topics in email
    • One panel during dinner friday evening and one panel during mid-morning saturday
    • Brief intro to panel before we go to dinner
    • Vu-graphs didn't generate much discussion (most of Q&A went in other directions)
    • May have some time to break into dinner room (from 4:45 to 5:00 to do something before adjourning for dinner)
    • Mark to start email discussion to create 5 seed questions for each panel
    • Lois to start email discussion on panelists for each panel
  • Most recent agenda
    • Repo has the most recent agenda
    • Lori usually introduces this at end of intro talk
    • FASTMath tutorial goal(s)
      • Will continue
      • Lori slotted to speak about algebraic solves and FASTMath (general intro to projects not being presented)
    • What about timing (some hour slots, some 45 min. slots and some 30 min slots). Is this all rigid?
  • Should we meet another time? August 2nd?
    • Will plan to me August 2nd but may cancel if not needed

July 5th, 2016

  • Attendees: Lois, Cameron, Mark M., Glen, Rob F., Mark S.
  • Agenda Items:
    • Software Installs
    • Panel Topics
  • Software Installs
    • Whats the status?
      • The existing ATPESC_2016 dir is bogus, a left over install from Mark M's student from earlier this year. Mark will clean it out.
      • Mark to start installs mid-week this week.
    • Need to contact package managers for desired versions
      • What version of Hypre (Rob F. was available on the telecon to ask)? 2.11.1
    • Mauro Pegra will work with Glen on ice-sheet problem/exercise setup
    • Cameron/Glen will work on Pumi/Albany installation
    • Pumi/Albany will likely install its own low-level infrastructure like pnetCDF, HDF5, etc.
    • Can we leverage trilinos install for other Sandia packages (nox, ml, zoltan)
      • Lets start an email discussion about that
      • Could run nox test problems too
    • Can we use xSDK toolkit installer will install many of FASTMath packages?
      • At some point, it woud be good to maybe exericse this and send feedback to IDEAS team.
  • Panel Topics:
    • Add a focus area that is related to progress being made on existing MC/GPU architectures
    • Start panel off by giving an update on the machines being installed
      • Maybe students will already been exposed to this but summary here may be relevant
    • Experiences in moving (scaling) in these directions (new archs) with these specific tools
      • For example, unstructured mesh stuff and OO design with pure MPI, but with MC systems had to develop array-based procedures, gpu=more restrictive array based procedures
      • How does threading really effect this? How does MC vs. GPU threading really work?
        • Barry can participate friday but not saturday
    • Note: One Panel during Friday dinner and one during Saturday
      • Dinner extended 1/2 hour to accomodate need to panel/speakers to eat before participating in panel
  • Action Items
    • Mark
      • Email package managers for versions to install
      • Clean out existing install point
      • Test running some parallel jobs/exercises
      • Startup discussion off-line about installation sharing
      • Craft panel invite on adjustments to topic areas and speakers
        • If there are things that work for panelists or to suggest other adjustments to the topics
    • All speakers
      • 3-slide overviews in by July 19th
      • first draft presentations by July 26th
      • Agree to convene this meeting again on 19th and 26th

May 10, 2016

  • Attendees: Mark M., Brian, Cameron, Lois, Rob, Carol, Vijay, Lori, Mark S.
  • Agenda Items
    • Everyone registered?
    • Schedule changes?
      • Draft Agenda
      • Sherry's inquiry?
      • Will need to figure out how to shift things around. Will work that out with Sherry
      • Cannot cut down on break, lunch or later in the evening
  • Dinner speaker has been selected
  • Lori's comments on funding for meeting
    • Everyone was in relatively good shape
    • May get an extension of all institutes until end of CY (up to 4 months)
    • Not yet confirmed but strong rumor
  • Panel topics:
    • Performance portability
    • Horror stories with extreme scaling
      • Maybe refine this for some fresher stuff to talk about for this
      • Extreme scaling in the context of these different architectures
        • Too close to PP topic?
        • What are some of the things you are looking at to date for heterogeneous environments
        • First crack, we don't worry about PP. But, now, re-evaluate for PP
    • What does it mean to deal with these heterogeneous platforms?
      • On-node vs. cross-node operations
    • Types and hierarchies of memory, memory footprints, performance models
      • Hetero compute, execute and data placement
    • Questions for students
      • Will students be told arch. or aurora and summit.
      • Will have to look at ATPESC agenda to confirm but likely students will have familiarity with these architectures by the time FASTMath session begins
      • May need just a few slides introducing/describing these architectures to frame the panel discussion

March 1, 2016

  • Attendees: Mark M., Cameron, Lois, Brian, Carol, Lori, Rob, Glen, Sherry, Mark S., Vijay
  • July 31 - Aug 12
  • FASTMath is August 5th and 6th
  • FASTMath has been asked to confirm draft agenda within next week online
    • Have some flexibility
  • Suggestions for dinner speakers
  • Time alotment expected to be most of friday and saturday morning
  • Jim Demmel may be part of friday morning
  • Travel same as before
    • Porgram will provide lodging, meals and limo to/from airport
    • Participants organization's need to cover airfare
  • Officially, FASTMath ends in August so need to pay attention to funding levels to support your participation at ATPESC
    • Not sure how other organizations are w.r.t. to funding
    • LBL is at zero already
      • Are there other funds that might be available to use for travel for this?
      • Lori to touch base with Esmond to check into situation at Berkeley
    • Is Sherry too? Not sure
    • RPI situation? Not positive what the end date is but not worried either
    • Sandia? Same boat as RPI. Don't see any problems to travel as planned. Will check for next call
  • Lori will talk with Bill H. about funding situation and when the new call will be out
  • Any proposed changes to agenda
    • Feedback from last time
      • Got very positive feedback about both panels. So, we should keep those for next time.
      • Most negative feedback is that FASTMath session is like drinking from a firehose. Should we consider what we attempt to teach in this time?
      • Any feedback on topics that people like
      • Does hands-on generate a lot of value?
        • Probably not allowed to cut this out but maybe need to continue to find way to improve how we execute the hands-on
        • Brian had experience of deep-dive on Chombo with a few
        • Lori would like speakers to maybe emphasize the ways of getting into contact
        • Could augment material with a sheet of contact lists for each package
    • Panel topics
      • Portable performance
      • Challenges going to extreme scale (or getting ready for Exascale)
      • Things library developers are doing for this generation and next generation of machines
        • Not so much the answers but sharing our experiences
        • Can even have some of the war stories of past experiences
  • FASTMath packages on ORNL
    • RPI no
    • MOAB no
    • PETSc? yes
    • Chombo? yes
    • SuperLU? did testing there and performance
    • Albany? yes, nightly testing, all trilinos, zoltan, nox, ml
    • Hypre?
    • Sundials? no

ATPESC_2016_Agenda_2016-08-5_8.pdf - Draft Agenda for FASTMath Session (45.4 KB) Mark Miller, 05/11/2016 01:07 am