buildworld in 45 min!

[Valeri Galtsev]

On Thu, February 5, 2015 4:39 pm, Bob Willcox wrote:
> On Thu, Feb 05, 2015 at 04:32:39PM -0600, Valeri Galtsev wrote:
>>
>> On Thu, February 5, 2015 3:51 pm, Bob Willcox wrote:
>> > On Thu, Jan 08, 2015 at 03:01:41AM -0800, Anton Shterenlikht wrote:
>> >> I got a Sun Fire X4100 for 60 pounds off ebay:
>> >>
>> >> FreeBSD 10.1-RELEASE #0 r274401: Tue Nov 11 21:02:49 UTC 2014
>> >> CPU: Dual-Core AMD Opteron(tm) Processor 2220 (2792.64-MHz K8-class
>> CPU)
>> >> real memory  = 17179869184 (16384 MB)
>> >> avail memory = 16595623936 (15826 MB)
>> >> FreeBSD/SMP: Multiprocessor System Detected: 4 CPUs
>> >> FreeBSD/SMP: 2 package(s) x 2 core(s)
>> >>
>> >> It does buildworld -j8 in under 45 min.
>> >> That's really fast!
>> >> I'm used to 4 hour ia64 builds.
>> >> Time to move on perhaps...
>> >>
>> >> Anton
>> >
>> > Not trying to show anyone up (really), but you can just imagine how
>> > shocked I
>> > was when I first ran a buildworld on my current workstation system in
>> 11
>> > minutes!
>> >
>> > FreeBSD 10.1-STABLE #3 r277033: Sun Jan 11 17:07:24 CST 2015
>> > CPU: Intel(R) Core(TM) i7-5820K CPU @ 3.30GHz (3300.07-MHz K8-class
>> CPU)
>> > real memory  = 34359738368 (32768 MB)
>> > avail memory = 33184337920 (31647 MB)
>> > FreeBSD/SMP: Multiprocessor System Detected: 12 CPUs
>> > FreeBSD/SMP: 1 package(s) x 6 core(s) x 2 SMT threads
>> >
>> > The latest run was 11:43 (looks like I'm losing a little ground). This
>> is
>> > with -j12.
>>
>> I couple of times had notices incomplete build with -j16 (or whatever
>> those boxes allowed), ever since run buildworld as a single thread...
>> Maybe just me ;-)
>>
>> Valeri
>
> Hmm, I have had quite some success with parallel builds myself. However, I
> never bother to go beyond the number of logical CPUs in the system.

So do I. "16" was a figure of speech. I usually would use number of CPUs
less by 2 than number of physical CPUs on the box (leaving a couple of
CPUs for the rest that runs on the box). Otherwise you loose on process
switching overhead between processes sharing the same CPU. Not only 0 gain
in speed, but loss on process switching overhead.

As far as incomplete builds are concerned, I indeed have seen them
occasionally, so, lazy person's solution was: single thread build ever
since ;-) Again, probably just me. And probably at particular point in
time with particular compiler...

Valeri

> When I
> did
> try that in the past it didn't seem to really speed things up all that
> much
> (if any).
>
> Bob
>
>>
>> >
>> > I can remember building 386bsd back in the early '90s on a 386 system
>> and
>> > it taking
>> > just under 24 hours!  :)
>> >
>>
>>
>> ++++++++++++++++++++++++++++++++++++++++
>> Valeri Galtsev
>> Sr System Administrator
>> Department of Astronomy and Astrophysics
>> Kavli Institute for Cosmological Physics
>> University of Chicago
>> Phone: 773-702-4247
>> ++++++++++++++++++++++++++++++++++++++++
>
> --
> Bob Willcox    | You climb to reach the summit, but once
> bob at immure.com | there, discover that all roads lead down.
> Austin, TX     |       -- Stanislaw Lem, "The Cyberiad"
> _______________________________________________
> freebsd-questions at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-questions
> To unsubscribe, send any mail to
> "freebsd-questions-unsubscribe at freebsd.org"
>


++++++++++++++++++++++++++++++++++++++++
Valeri Galtsev
Sr System Administrator
Department of Astronomy and Astrophysics
Kavli Institute for Cosmological Physics
University of Chicago
Phone: 773-702-4247
++++++++++++++++++++++++++++++++++++++++