[ANN] EmacsConf2023 presentation with analysis of the collected statistics

[Ihor Radchenko]

Dear all,

Thanks everyone who submitted the data!

I have now summarized everything.
You can find the analysis in my EmacsConf2023 presentation -
https://emacsconf.org/2023/talks/gc/ and
https://dx.doi.org/10.5281/zenodo.10213384 (reproducible analysis).

The key takeaways from the above analysis:

  For users:
  1. Emacs freezes due to GC are real
  2. … but not for all users
  3. When GC does get Emacs slower, it is either
     • slow startup ⇒ increase `gc-cons-threshold' to 20-80Mb check
        `gc-elapsed' after startup
     • slow and frequent GC ⇒ increase `gc-cons-percetage' to 0.2-1.0
        but watch Emacs RAM usage

  For Emacs devs:
  1. GC problem is very real for many users (9% of GCs are both frequent
     and slow)
  2. It is worth bumping `gc-cons-threshold' temporarily during startup
  3. Ideally, we should switch from Mark&Sweep to something more
     performant
  4. May increase `gc-cons-percetage', but need to get user data for
     optimal numbers

Further plans:

1. With the collected data we can hopefully continue emacs-devel
   discussion on changing GC defaults and get somewhere.
   https://yhetil.org/emacs-devel/20230310110747.4hytasakomvdyf7i@Ergus/

2. emacs-gc-stats package has now served its purpose, but I think that
   it might still be useful for people who want to toy around GC
   settings. I plan to add some kind of "session summary" command that
   will display the key statistics of current user session. It should
   help understanding the effects of GC customization on the performance.

-- 
Ihor Radchenko // yantar92,
Org mode contributor,
Learn more about Org mode at <https://orgmode.org/>.
Support Org development at <https://liberapay.com/org-mode>,
or support my work at <https://liberapay.com/yantar92>