This repository has been archived by the owner on Oct 23, 2020. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 0
Agenda Notes Monday December 18th 2017 11:00 AM
Michael Duda edited this page Dec 18, 2017
·
7 revisions
Date: 18 December 2017
Time: 11:00 MDT / 19:00 CET
Call-in number: 1-866-740-1260
Access Code: 4978161
- Framework modifications:
- Discuss E3SM/MPAS repo set-up
- Short report on MPAS-A speed-ups when switching to grouped halo exchanges
Version 36 72 144 288 576 1152 2304
v5.2-opt 567.36 268.62 123.54 57.21 27.72 13.77 7.1
grouped-halos 562.15 266.27 123.34 58.44 26.98 13.65 6.92
Monday, 11 December 2017 @ 11:00 MDT
Participants: Adrian, Matt, Mark, Michael
- MPAS-Atmosphere timings for the dynamics, which have essentially all of the halo exchanges in MPAS-A, are show in Item 3 of the agenda. These numbers are from global simulations on the x1.40962 mesh on NCAR's Cheyenne system using the Intel compilers and SGI MPT. There doesn't seem to be a clear speed-up when grouping halo exchanges.
- Depending on machine, MPI library, etc., perhaps grouped halo exchanges need more exchanges to be grouped together to show large improvements?
- Adrian will soon have performance measurements with grouped halo exchanges for MPAS-Seaice when using up to ~20,000 cores of Cori (or Edison?)
- E3SM will make use of a completely open development model
- How might we handle the MPAS repository and its interaction with the E3SM repository?
- We will probably need recommendations in the January - February time frame
- We'll start a brainstorming document (Google Doc) to discuss pros/cons of options for repo management
- Plan for a high-level telecon on 29 January