2018 08 07

   - TODO T1: launch methods need to be tested
     - tests added for major LMs
     - more to be done about the scheduler data structure
   - TODO T2: check BW ramdisk under ORTE
     - depends on #1686
   - Experiments
     - Jumana
       - surprising reduction also in scheduling time
     - Srinivas
       - as expected
   - declare success
     - end-2-end impl should come earlier
     - managed to be not too generic
     - use case driven: ok
     - tacit assumptions are tricky
       - discuss in groups
       - test early
     - feedback to us (boring / useless / ineffective)?
       - design / exploration: in whole class, not separate classes
         - reduces later friction
         - shared assumptions
         - 1:1 syncs helped (VB, IP)
         - group structure / hierarchy always emerges -> (self)-organization
           - this is part of the experience
       - working out of my area of expertise
         - that's the point :-)
       - workload too much?
         - VB: not because of the topic, just finding the time every week
       - match to technical skills (RP, coding, project management)?
         - steep curve - in a good way
       - did we miss anything technical?
         - probably not?
     - we will repeat the course :-)
       - passing means moving on to development