[distcc] distcc gives sixfold speedup
tim at digitalbrain.com
Mon Sep 30 12:33:00 GMT 2002
Martin Pool wrote:
>>6. Any other observations:
>> If (say, due to a reboot) distccd isn't running on the distccd
>> servers, "make -j 24" ends up spawning 24 copies of g++ on the
>> build master, which harms interactive response, to put it
>> lightly. (it sometimes takes a few 10s of seconds to get control
>> back to kill the make)
This was an issue with our old perl+ssh+lvs based distribution scheme.
What I did was add a switch to the perl script, so that it returned 0
or 1 depending on whether distributed compiles were "available" in a
general sort of way. Our build scheme used this, and then set -j 8 or
-j 1, depending on what it got back...
It was on my TODO list to get it return a general "this is how many CPUs
I have available" type number, but I never got around to it ;-).
More information about the distcc