[Nix-dev] Re: Performance impact of Nix

[Marc Weber]

Hi Tony,

you're aware that the system caches directory contents?
So if you run the same command a second time everything will be faster?

So what are you measuring exactly ? The time to of man 3 glob ?

 marc at nixos: /nix/store ]$ time  echo *d5rw61pr6ffl0mnbrfs2*
32d5rw61pr6ffl0mnbrfs2ih5xjk4nls-vim_configurable-7.2

real    0m0.402s
user    0m0.380s
sys     0m0.020s
 marc at nixos: /nix/store ]$
marc at nixos: /nix/store ]$ time echo
32d5rw61pr6ffl0mnbrfs2ih5xjk4nls-vim_configurable-7.2/*
32d5rw61pr6ffl0mnbrfs2ih5xjk4nls-vim_configurable-7.2/bin
32d5rw61pr6ffl0mnbrfs2ih5xjk4nls-vim_configurable-7.2/share

real    0m0.000s
user    0m0.000s
sys     0m0.000s


So the command using glob is much slower. So time also measures the
globbing time!

I  mean all I saw is:
> real    0m0.065s < first run
> real    0m0.058s < next runs maybe using cached stuff..
> real    0m0.055s
> real    0m0.059s

So I wouldn't conclude that that's the difference unless you retry many
times.

Glob will always read all store paths and test the glob pattern.
If you test for *foo* it doesn't actually matter where this *foo*
appears that much if it's hit only once.. because foo hast to be matched
again and again anyway.

/nix/store/EEEEEEEEEEEEEEEEEEEEEEEEEEEEe-name
           foo* ? no
            foo*? no
             foo*? no
            ....
Of course this can be optimized but I don't think that this changes glob
time behaviour too much.

I think the more important question is: I accesing
  /EEEEEEEEEEEEEEEEEEEE-name

faster than
  /nix/store/EEEEEEEEEEEEEEEEEEEEEEEEEE-name 
or
  /nix/store/a/b/c/d/e/f/EEEEEEEEEEEEEEEEEEEEEEEEEE-name 
?

And I don't think you can measure this using one directory access only
using the time command because process scheduling will modify your
results as well then.

Marc Weber