2021.09.02 Meeting Notes

Agenda

• Individual/group updates
• Paper
• Review non-WIP PRs
• Next release

Individual/group updates

LANL/External CS

Tanner Broaddus

• Busy with classes, starting full-fledged next week

Jonas Lippuner

• It's working without AMR on single node (no MPI)
• Running but not producing correct results yet since with AMR, issue is that flux corrections are not triggering allocation
• Will push current work to WIP PR
• Question: Should we make sure true sparse does exactly the same as when variables were dense, or would we be ok with some inaccuracies if it makes implementation easier (e.g. only checking for new allocations at end of all stages)

LANL Physics

Jonah Miller

• Preparing submission for IC allocation on new A100 machine, other Parthenon developers would be able to use allocation. Question: could we run a whole-machine scaling test before paper submission? If not, maybe before referee response.
• Doing some scaling and performance tests for this proposal, seeing basically perfect scaling
• Played with some A100 GPUs for Phoebus (downstream code), seeing good performance

Ben Ryan

• Working with Jonah on the IC proposal, but couldn't get CUDA aware MPI support, trying to figure that out
• Polishing the particles boundaries PR, should be mostly ready now

AthenaPK

Philipp Grete

• Baby is here and healthy
• Reviewed Ben's particle boundaries PR
• Fixed some logic issues in output
• Will reintroduce signaling based output (with final in name instead of id), currently it seems that if a signal is caught, no final output is produced

Forrest Glines

• Working on PR to facilitate some analysis
• Focusing on AthenaPK

Jim Stone

• Some edits on paper

Paper

• Paper seems to be in better shape than last time
• Philipp will have some more time to write soon-ish
• Philipp: Are Phoebus and RIOT read to produce publication-ready performance metrics/figures?
• Jonah: Phoebus has pretty good scaling results right now, RIOT may not be quite there yet
• Focus on raw zone-cycles/s and look at meshblock sizes
• Jim: Showcase different types of physics that can be implemented with Parthenon using the different downstream codes and compare performance to other published results (as opposed to just showing hydro in different downstream codes)
• Agreement on Jim's suggestion. Might even talk about Ben's Monte Carlo stuff in Parthenon
• Homework: decide on what plots to show for downstream codes

Release

• Philipp: Will take care of next release after Ben's particle boundary PR is merged