Meeting minutes – Feb 20, 2025 (12PM ET)
Present: Cheng-Hsin, Erik, Jay, Liwei, Lucas, Roland, Swapnil, Yosef, Zachariah
Roland: *Merged the pull request for symmetry BCs in NewRadX. Test the thorn. *Subcycling mostly done except for writing the paper. See how to get these to main branch -> would be good to have multiple commits with separate feature implementations. Liwei says it is already the case. *Roland asks to scrutinize the optimizations to make sure they are safe. *Roland commented that Liwei's pull request looks gigantic -> Liwei says it is due to unit tests. *Roland trying to craft test cases for Z4c to include in ET. Will try and see if same ones as McLachlan can be used. Also a qc0 test in Z4c that curretly fails. Some work that needs to be done for subsyling -> Granularity. Ronald is not running Z4c from spacetimeX (which works), but directly from ET.
Cheng-Hsin: Could not provide updates due to audio issues.
Zachariah: *Roland says try and see if BHaHAHA thorn works with CarpetX. Zach testing it. *In the BBH mode -> finds specific horizons and when BHs get closer, starts searching for a common horizon and disables searching for different horizons. Seems to be working correctly.
Lucas: *Roland asks Lucas about the curvilinear coordinate work and Hybrid time integrators. *Lucas says that Multipatch seems to be working correctly in CapyrX. Successfully ran qc0 and extracted multipole. Ran for 200-300 timesteps, need to run longer for useful results. Interpatch communication seems to be working. Exciting! *There is some trouble with openpmd output. CPU clusters with spack give warning related to BP5 and produce corrupted files. Suspect something wrong with adios that comes with spack. Try to use Silo, Erik will look into it. *No big updates on hybrid time integrators. Currently, tuning methods to maximize CFL while keeping acceptable results.
Jay: *Doing scaling test with asterX on frontier using subcyling. Sometimes get output (e.g cell updates per second) which is not consistent across different iterations. Even an order of magnitude difference at times. *Could be due to processor mismatch. Roland suggests using barrier at the end of every timestep. Erik says carpetX already has barriers at the end of every timestep, and function in schedule.cxx does MPI_AllReduce which is a global reduction. Outputs in the evolution loop:
CCTK_VINFO("Simulation time: %g "
"Iterations per second: %g " "Simulation time per second: %g",
*Erik suggests to use parameter out_performance. Discussed the code further.
*Roland updated on the UIUC workshop 1st-2nd week of August.
From: Carpetx-developers carpetx-developers-bounces@einsteintoolkit.org on behalf of rhaas@illinois.edu rhaas@illinois.edu Date: Wednesday, February 19, 2025 at 5:08 PM To: carpetx-developers@lists.einsteintoolkit.org carpetx-developers@lists.einsteintoolkit.org, rhaas80@gmail.com rhaas80@gmail.com Subject: [Carpetx-developers] Weekly call at 12pm Thursday Eastern time Title: CarpetX developer call When: 12pm Thursday Eastern time
Join Zoom Meeting https://eur04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpitp.zoom....https://pitp.zoom.us/j/93861697232
Join by phone (CA) +1 647-558-0588
Join using SIP 93861697232@zoomcrc.com _______________________________________________ Carpetx-developers mailing list Carpetx-developers@einsteintoolkit.org https://eur04.safelinks.protection.outlook.com/?url=http%3A%2F%2Flists.einst...http://lists.einsteintoolkit.org/mailman/listinfo/carpetx-developers