BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20190719T085744Z
LOCATION:HG D 1.2
DTSTART;TZID=Europe/Stockholm:20190613T164500
DTEND;TZID=Europe/Stockholm:20190613T171500
UID:submissions.pasc-conference.org_PASC19_sess168_msa128@linklings.com
SUMMARY:Experiences Refactoring the Energy Exascale Earth System Model – M
 ultiscale Modeling Framework (E3SM-MMF) for the Summit Supercomputer
DESCRIPTION:Minisymposium\nComputer Science and Applied Mathematics, Clima
 te and Weather, Physics\n\nExperiences Refactoring the Energy Exascale Ear
 th System Model – Multiscale Modeling Framework (E3SM-MMF) for the Summit 
 Supercomputer\n\nNorman\n\nThe E3SM-MMF is a compromise between traditiona
 l climate simulation and explicit global cloud resolution, wherein every G
 lobal Climate Model (GCM) column has its own embedded Cloud Resolving Mode
 l (CRM) to explicitly simulate clouds on a reduced domain. The CRMs at dif
 ferent GCM columns do not communicate with one another directly. Rather, t
 hey communicate via the GCM itself. In the MMF setup, each CRM’s ave
 rage is forced to match the GCM column data, keeping each GCM column tight
 ly coupled with its embedded CRM. The CRM, in turn, gives the GCM column t
 ime tendencies of certain cloud properties. This talk details experiences 
 refactoring the CRM code to utilize GPUs on the Oak Ridge Leadership Compu
 ting Facility’s (OLCF’s) Summit supercomputer. Given this is a
  large, actively developed Fortran code, we use OpenACC directives. Climat
 e is a unique use case in GPU porting because the 2,000x realtime throughp
 ut requirement requires strong scaling to the point that very little work 
 is available per node. This means all available threading must be exposed,
  leading to unique refactoring approaches. While results are presented, mo
 re emphasis is placed on the refactoring process itself in the hopes of ge
 nerating discussion toward finding better software engineering practices.
END:VEVENT
END:VCALENDAR

