Performance and profiling for imaging

[peijin94]

Profiling tools : line_profiler(pip install), cProfile(python built-in), snakevis(pip install)

The following is the time consumption breakdown for the pipeline

image_ms_quick

• 16% bandpass
• 50% selfcal
• 25% source removal

do_selfcal

• 50% gaincal
• 37% wsclean call

source_subtraction:remove_nonsolar_sources

• 84% wsclean
• 10% predict (de-gridding)

Ideas to improve

○ alternative gaincal in casa (takes 25% overall time) ✅ remove_nonsolar_sources use allsky image from the previous selfcal step. (will save 15-20% of overall time) ○ something about bandpass?

[peijin94]

auto_pix_fov=True will bring 10% overall performance gain (147s to 132s). The all-sky image size becomes 2400 at 55MHz, (by default 4096).