Re: How to improve Org startup time?

[Nick Dokos <nicholas.dokos@hp.com>]

Sebastien Vauban <wxhgmqzgwmuf@spammotel.com> wrote:

> >> This may have something to do with my big amount of Org files in
> >> `org-agenda-files': 36 at this point. But is that so big??
> >
> > I don't think so.
> 
> I'm sure it is, as I wrote (removed from the joined log) a "message" call
> before and after that line, and those time-stamps were 16 seconds one from the
> other.
> 

Small misunderstanding here: the "I don't think so" was a reply
to the "is that so big?" question.

> > Can you get a profile?
> 
> You're right I should use real tools for that...
> 
> > M-x elp-instrument-package RET org-agenda RET
> > M-x org-agenda-to-appt
> > M-x elp-results
> >
> > In my (admittedly unchallenging, run-of-the-mill) setup, I get
> > (with everything already loaded):
> >
> > org-agenda-to-appt                                            1           0.053846964   0.053846964
> > org-agenda-prepare-buffers                                    1           0.028483817   0.028483817
> > org-agenda-get-day-entries                                    8           0.024222044   0.0030277555
> > org-agenda-get-scheduled                                      8           0.0154506449  0.0019313306
> > org-agenda-get-timestamps                                     8           0.004179949   0.0005224936
> > org-agenda-skip                                               184         0.0027937810  1.518...e-05
> > org-agenda-files                                              2           0.001386068   0.000693034
> > org-agenda-get-deadlines                                      8           0.001288303   0.0001610378
> > org-agenda-format-item                                        22          0.0012140690  5.518...e-05
> > org-agenda-get-blocks                                         8           0.0007851970  9.814...e-05
> > org-agenda-new-marker                                         44          0.0006114019  1.389...e-05
> > org-agenda-skip-eval                                          368         0.0001511950  4.108...e-07
> > org-agenda-todayp                                             16          0.0001342800  8.392...e-06
> > org-agenda-fix-displayed-tags                                 22          7.2914e-05    3.314...e-06
> > org-agenda-get-category-icon                                  22          1.8382e-05    8.355...e-07
> > org-agenda-time-of-day-to-ampm-maybe                          6           3.347e-06     5.578...e-07
> >
> > A factor of 300: maybe it's real, but let's make sure first.
>
> Here it is. I don't know how to interpret that difference, tho.
> 
> org-agenda-to-appt                                            1           19.673        19.673
> org-agenda-prepare-buffers                                    1           18.86         18.86
> org-agenda-get-day-entries                                    36          0.7970000000  0.0221388888
> org-agenda-files                                              37          0.5440000000  0.0147027027
> org-agenda-get-scheduled                                      36          0.5150000000  0.0143055555
> org-agenda-get-deadlines                                      36          0.1580000000  0.0043888888
> org-agenda-skip                                               612         0.1410000000  0.0002303921
> org-agenda-get-timestamps                                     36          0.047         0.0013055555
> org-agenda-get-blocks                                         36          0.046         0.0012777777
> org-agenda-format-item                                        42          0.031         0.0007380952
> org-agenda-skip-eval                                          1204        0.016         1.32...e-005
> org-agenda-fix-displayed-tags                                 42          0.0           0.0
> org-agenda-todayp                                             72          0.0           0.0
> org-agenda-new-marker                                         89          0.0           0.0
> org-agenda-deadline-face                                      2           0.0           0.0
> org-agenda-get-category-icon                                  42          0.0           0.0
> 

Well, you have a bigger agenda by a factor of 4-5 and I guess a slower
machine, but it all takes less than a second except for one thing: the
big difference seems to be org-agend-prepare-buffers which
opens the files, reads them in and gets the buffers ready. 
Once the files are open however, they should be cached, so doing it
again should not take very long at all: is that the case?

If so, my guess is that either you have a very slow disk or you have
disk problems. Were things fast at some point in the past? In that case,
I would worry about the disk. Also, what happens if you copy your agenda
files to some other machine and try it there? If that's fast, then again
I would worry about this disk.

An anecdote to illustrate: at one point, it would take me a couple of
minutes to log in whereas previously it was a few seconds.  The kernel
log contained a few disk errors. The disk errors apparently caused multiple
retries, both in the kernel but also (and most probably more
time-consumingly) in the disk firmware. The errors all happened to be in
the file that contained my desktop background image. When I changed
backgrounds (leaving the old file in place so that the bad spots would
not be used by some other file), my login time went back to a few
seconds. I replaced the disk and (knock on wood) the problem has not
reappeared.

So check your kernel logs and maybe run a disk diagnostic program as well.
If there are errors, try backing up the agenda files (copying them
should take a long time if my suspicion is correct): do the equivalent
of

	mv file1.org file1.org.bak
	# the following might take a long time
	cp file1.org.bak file1.org

so that the backups will still occupy the error loci. If the disk is
not too far gone, that might restore the speed of the agenda, but if
that's the case, don't wait too long to replace the disk: once a disk
goes a little bad, it tends to go a *lot* bad pretty fast.

Nick