Re: [O] How to improve Org startup time?

[Sebastien Vauban]

Hi Eric,

Eric S Fraga wrote:
> Some timings on a 3-4 (?) year old netbook with Intel N450 Atom @1.66GHz
> dual core powered:
>
> With old org, actually from October 2012 (!) as this system acts as a
> cloud server for me so I don't actually use it interactively much:
>
> | org-agenda-to-appt         |   1 | 12.525722775  | 12.525722775  |
> | org-agenda-prepare-buffers |   2 | 10.719484861  |  5.3597424305 |
> | org-agenda-get-day-entries |  15 |  1.4113181819 |  0.0940878787 |
> | org-agenda-get-scheduled   |  15 |  1.032943071  |  0.0688628714 |
> | org-agenda-get-deadlines   |  14 |  0.2688785840 |  0.0192056131 |
> | org-agenda-skip            | 797 |  0.1879489969 |  0.0002358205 |
>
> and with up to date org as of a few minutes ago:
>
> | org-agenda-to-appt         |   1 | 9.168879079  | 9.168879079  |
> | org-agenda-prepare-buffers |   2 | 8.1717637810 | 4.0858818905 |
> | org-agenda-get-day-entries |  15 | 0.5261351449 | 0.0350756763 |
> | org-agenda-get-deadlines   |  14 | 0.253867925  | 0.0181334232 |
> | org-agenda-get-scheduled   |  15 | 0.1835072029 | 0.0122338135 |
> | org-agenda-skip            | 797 | 0.0482942050 | 6.059...e-05 |
>
> I.e. an approximately 25% reduction in time overall so I think
> org-agenda is significantly faster (okay, not orders of magnitude) but
> the time taken is dominated by org-agenda-prepare-buffers on this
> system.
>
> However, more interestingly, on a 1 year old desktop computer Intel Core
> i5-2500 @ 3.30 GHz, I get very different behaviour:
>
> | org-agenda-to-appt         |   1 | 0.182718048  | 0.182718048  |
> | org-agenda-get-day-entries |  14 | 0.095832824  | 0.0068452017 |
> | org-agenda-prepare-buffers |   1 | 0.086104933  | 0.086104933  |
> | org-agenda-get-scheduled   |  14 | 0.067475949  | 0.0048197106 |
> | org-agenda-skip            | 794 | 0.0475198030 | 5.984...e-05 |
> | org-agenda-get-deadlines   |  14 | 0.014998552  | 0.0010713251 |
>
> Notice the big change in org-agenda-prepare-buffers relative to the
> other calls.  Disk performance is almost likely the key here and I think
> the recent improvements are quite dramatic but I haven't had a chance to
> time these yet.

Dividing the figures by 50, just for the disks, seems too big for me, nope?

Just to be sure: are you sure that, in this last test, `org-agenda-to-appt'
was called without having first opening the agenda files?

Best regards,
  Seb

-- 
Sebastien Vauban