benchmarks.txt

benchmarks.txt (text/plain), 2.65 KB, created by TGL on 2004-01-07 09:56:29 UTC

This "benchs" have been made on an Intel Centrino 1.4 GHz laptop, with 256 MB, 
running a 2.6.0-test11-mm1 kernel, with preempt and lowlatency enabled, and 
PORTAGE_NICENESS unset. For each package, I've done a first "emerge" which 
does not appear in this results, and then two verbose (normal) emerge, two 
"--quiet" emerges, and two "--quiet --nospinner" emerges, in random order.
This tests all include an unmerge (clean) of the previous package instance.

Results show a significant improvement of performance in quiet mode, especially
for very verbose merges like vanilla-sources (~300%). But it is also still 
significant for middle-size packages like ruby. That said, this results have
been obtained in gnome-terminal, which seems to be a slow terminal emulator.
Sure, there are absolutly no change in performance when using emerge in a
(non framebuffer) console, or redirecting the output to /dev/null. So this
quiet mode can be seen as a good workaround to "modern" terminal emulators :p
Also note that the tests with "--quiet --nospinner" don't show any significant
improvement compared to the simple "--quiet".

Something funny is that with PORTAGE_NICENESS=19, the verbose mode can be 
significantly faster. At first, since this was my config, I've spent quite a 
lot of time trying to understand where my patch was slowing things down. But 
this behavior can be reproduced with any non-patched portage by redirecting 
output to /dev/null. I think this is due to the kernel scheduling of what he 
sees as a more interactive task because of the dump it produces in the 
terminal, or something like this.


### app-pda/gtkpod ###

# verbose:
real	0m57.245s
user	0m38.781s
sys	0m5.514s
---
real	0m56.909s
user	0m38.916s
sys	0m5.467s

# quiet:
real	0m47.346s
user	0m38.801s
sys	0m5.501s
---
real	0m47.437s
user	0m38.806s
sys	0m5.491s

# quiet, nospinner:
real	0m46.722s
user	0m38.728s
sys	0m5.465s
---
real	0m47.079s
user	0m38.710s
sys	0m5.521s


### dev-lang/ruby ###

# verbose:
real    3m30.116s
user    1m54.074s
sys     0m21.492s
---
real    3m32.871s
user    1m53.159s
sys     0m21.906s

# quiet:
real    2m28.875s
user    1m52.836s
sys     0m21.874s
---
real    2m28.731s
user    1m52.225s
sys     0m21.160s

# quiet, nospinner:
real    2m29.831s
user    1m51.605s
sys     0m20.999s
---
real    2m28.263s
user    1m53.672s
sys     0m21.055s


### sys-kernel/vanilla-sources ###

# verbose:
real	10m14.438s
user	0m41.427s
sys	0m9.637s
---
real	9m25.832s
user	0m41.230s
sys	0m9.578s

# quiet:
real	3m15.261s
user	0m40.496s
sys	0m8.922s
---
real	3m22.542s
user	0m40.417s
sys	0m9.427s

# quiet, nospinner:
real	3m27.354s
user	0m40.103s
sys	0m8.507s
---
real	3m17.919s
user	0m40.146s
sys	0m8.699s