2021.09.16 Meeting Notes

Jump to bottom

Jonas Lippuner edited this page Sep 16, 2021 · 3 revisions

Agenda

Individual/group updates

LANL/External CS

Jonas Lippuner

Posted WIP PR for true sparse, working for non-MPI
Working on adding support for MPI

Tanner Broadus

Idea: do all-gather for load imbalance one cycle before load balancing
Caution: Is the all-gather actually a bottleneck for realistic simulations?
Request form Philipp: update profiling wiki page to add instructions to get MPI profiling trace

LANL Physics

Jonah Miller

Mostly working on downstream stuff
Talked with Tanner about MPI profiling, looks like all-gather and all-reduce may be bottlenecks

Ben Ryan

Working on particle boundaries, debugging some segfaults
On the calendar to put particle stuff into paper

AthenaPK

Philipp Grete

Boundary bug fix and flux_div PR now has test that we wanted, test now covers all boundary conditions
Also added check that a zero-initialized field doesn't suddenly have non-zero values due to uninitialized corners
Added some more text to paper

Forrest Glines

Trying to run galaxy cluster simulations, running into bugs or additional features needed for analysis
HDF5 segfaulting with compression, large file size, and deep AMR structure
Also working on yt frontend for Parthenon

HDF group results/reply

Happy to have them create a PR
In favor of run-time configurable options for HDF5
Jonah will reply

Paper

Philipp: Have data for buffer-in-one results, but will probably need arch-dependent loop-patterns for buffer filling before getting results
Ben and Jonah: will come up with test problems in downstream codes

Release

Waiting on bug fixes for particle communication issue