And Here I Thought buildworld/makeworld Was IO Bound

Tim Daneliuk tundra at tundraware.com
Tue Feb 11 00:05:42 UTC 2014


On 02/10/2014 09:03 AM, Warren Block wrote:
> On Sun, 9 Feb 2014, Tim Daneliuk wrote:
>
<SNIP>

>> So is the bounding function here actually CPU not IO?  Am I missing
>> something?
>
> Yes, it mostly is CPU.  Make sure you are running powerd on that i5, it enables turbo mode and gives a noticeable speed increase.

I enabled the various turbo features on the mobo and rebooted with powerd turned on.
It appears to knock about 2 minutes off the build, but I need a bigger baseline
than one try to confirm this.

>
> Also, consider using -DNO_CLEAN, leaving all the untouched object files in place for make to find and skip.  I've seen buildworld times under a minute with that on my i5 and SSD.  There are downsides (uname is not always updated correctly), but removing /usr/obj every so often and starting from scratch will fix that.

Right after I did a complete buildworld/kernel, I ran it again with this
option enabled.  Essentially, I wanted to see how long it takes
the make infrastructure to figure out it has nothing to do... about
2 minutes appears to be the answer.  If I were sure that everything
was being picked up properly (including uname), I'd do it this
way every time ;)

Thanks for the feedback and help.

-- 
----------------------------------------------------------------------------
Tim Daneliuk     tundra at tundraware.com
PGP Key:         http://www.tundraware.com/PGP/



More information about the freebsd-stable mailing list