I am attempting to use esmf-trace to plot the traces generated by ESMF for a OM3 25k IAF experiment. In doing so, I see that we need to add more user documentation with an example of how to go about i) generating the traces (already done in the repo README) ii) installing the necessary tools and post-processing the traces (and choices that need to be made - e.g., can all trace streams be combined, should we pick out some streams that might be interesting) iii) displaying + saving the results, and iv) interpreting the results (what is the viz saying, what should we focus on etc)
The 25k scaling notebook here is a good starting point, but we also need to document how to inspect the timeline and optimise the i) partitioning to improve load-balancing, and ii) reorder the run sequence for a given partitioning
Noting to keep track for future efforts; I will attempt to update with my findings on this issue - so that can feed into future docs
I am attempting to use esmf-trace to plot the traces generated by ESMF for a OM3 25k IAF experiment. In doing so, I see that we need to add more user documentation with an example of how to go about i) generating the traces (already done in the repo README) ii) installing the necessary tools and post-processing the traces (and choices that need to be made - e.g., can all trace streams be combined, should we pick out some streams that might be interesting) iii) displaying + saving the results, and iv) interpreting the results (what is the viz saying, what should we focus on etc)
The 25k scaling notebook here is a good starting point, but we also need to document how to inspect the timeline and optimise the i) partitioning to improve load-balancing, and ii) reorder the run sequence for a given partitioning
Noting to keep track for future efforts; I will attempt to update with my findings on this issue - so that can feed into future docs