Optimizing hydro zooms for MPI
Follow-up from an e-mail discussion on optimizing zoom simulations over MPI, including tasks handling and domain decomposition.
A few items to discuss could be:
- Recap of the updates in Will’s dev branch.
- Creating a test suite comparing the runtime/profiling of the dev and master branches.
I could adapt one of my smallest zoom ICs, but this took 10.7 hrs on single Cosma7 node:
/cosma/home/dp004/dc-alta2/data6/xl-zooms/hydro/L0300N0564_VR3032_-8res_MinimumDistance_fixedAGNdT8.5_Nheat1_SNnobirth/
We might want ICs that run 3x quicker or end the simulation at e.g. z=2. - Splitting the work among us and next steps to take.