And Here I Thought buildworld/makeworld Was IO Bound

[Glen Barber]

On Mon, Feb 10, 2014 at 02:47:59PM -0600, Tim Daneliuk wrote:
> On 02/10/2014 02:43 PM, Glen Barber wrote:
> >On Mon, Feb 10, 2014 at 02:40:09PM -0600, Tim Daneliuk wrote:
> >>On 02/10/2014 10:07 AM, Glen Barber wrote:
> >><SNIP>
> >>
> >>>
> >>>This machine is 48-core (4 x 12 cores) Opteron 6174 (2.2GHz), 128GB RAM,
> >>>with 5 drives in a raidz1 on a PERC H700 controller.
> >>>
> >>>Glen
> >>>
> >>
> >>So ... in that config, you're saying that a buildworld/kernel time
> >>on the order of 30-ish minutes is about right, is that correct?
> >>
> >
> >Yes, for a stock (GENERIC kernel, empty src.conf and make.conf), this is
> >correct.
> >
> >Glen
> >
> 
> 
> 'Really interesting given that I am getting comparable times
> on a lowly quad I5 with 8GB.  It speaks pretty loudly
> to just how much you can scale up the compilation.  Of course,
> I'd expect your boxes to be able to do more of them simultaneously
> (scale out)...
> 

There are two things to consider though.  I'm not cranking the number of
'-j' as high as it probably can go.  I'm intentionally limiting it to
'j10' and '-j6' in order to do parallel builds.  I'm sure '-j24' or
higher in this case would reduce overall build time.

I'm just not comfortable doing that, because there are a few (sometimes
easily-triggered, other times not so easily-triggered) race conditions
that can cause the builds to fall over to easily.  So limiting to
a relatively safe number, scaling out certainly does help.  But this is
a very specific, and probably atypical use case. :)

Glen

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 834 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20140210/8e00b87f/attachment.sig>