[U-Boot] [PATCH 2/2] WIP: Test version of buildman - U-Boot builder

[Simon Glass]

Hi,

On Thu, Nov 1, 2012 at 4:29 PM, Tom Rini <trini at ti.com> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 11/01/12 14:01, Simon Glass wrote:
>> Hi Andy / Tom,
>>
>> On Wed, Oct 31, 2012 at 9:17 PM, Andy Fleming <afleming at gmail.com>
>> wrote:
>>> On Wed, Oct 31, 2012 at 8:18 PM, Tom Rini <trini at ti.com> wrote:
>>>> On Wed, Oct 31, 2012 at 05:11:51PM -0700, Simon Glass wrote:
>>>>
>>>> [snip]
>>>>> Also rather than running with 'make -j40' or whatever, it
>>>>> uses 'make -j1' but with 40 threads. This increases CPU
>>>>> utilisation quite substantially (almost 50%) - I think this
>>>>> was discussed some time ago,
>>>>
>>>> MAKEALL supports this, but doesn't default to it, today.  I'm
>>>> wondering if we ought to make it the default for all non-single
>>>> board builds.  The winning point is right around
>>>> boards-to-build == `grep -c processor /proc/cpuinfo` and I
>>>> think is right around setting NBUILDS to that value, both on
>>>> consumer multicore and heavy-duty 32/64 core boxes.
>>>
>>> Yeah, I usually set BUILD_NBUILDS to 24-50 on my 24-thread
>>> system, and then set BUILD_NCPUS to 1-4. It's nice to offer a
>>> little bit of parallelism on the individual builds, sometimes. I
>>> did a bunch of tests, but not in a scientific fashion. Enough to
>>> agree with you that maxing out builds seems to win for u-boot.
>>
>> As a bit of an unscientific test, what sort of time does it take
>> to build all 1000-or-so boards on your systems?
>
> My setup using MAKEALL does 921 boards (arm/powerpc/mips) with ELDK
> 5.2 in 54min wall-clock with 397% CPU util (on a 6 core machine).

buildman is best at building a series of commits at once. I suspect
with just one commit its performance would be similar to MAKEALL. I
use it to verify a series so that I know it will bisect correctly.

Has anyone given it a try? I have found one bug where it gets confused
when a builder thread moved from one architecture to another. Apart
from that I am thinking of cleaning up the series and submitting some
proper patches. Any comments?

Regards,
Simon

>
> - --
> Tom
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.11 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://www.enigmail.net/
>
> iQIcBAEBAgAGBQJQkwXKAAoJENk4IS6UOR1Ws+oP/A2Estr8LHk2zt3cclDrshwI
> 9iNpiMvwJ2a/Qj3DzK6q3tuLKIJq/Chfsgczrz3wlv6RGqN5SB8U/Cbl9/OAg9oC
> jmSEjpGxT3pJyRKscyf/78cjMb06BadHVybPjkMtmfdlay3+H4CLLFkVahM1BKut
> +7MxTFtQIkPkHtXUwcTSYXuIhwwz61d6R65colHoHWVIbtRWpksaWa64107mEg3K
> 0A4Ite+FYoMK0BeswA1mPGknDCCLfgE32oXI4k4tQBAXrLcYWKlxLcirGydiZ46l
> Mrqh+62EA4U2FEVj3VbRjB2JtqRiqtbYYNSk9fwkdU0Fa/eYZacYYxStZVnVTfaQ
> RgZd0Gq7knJUed8bS0RvptDKLnMCadc0CgVEQ6Svj2JgfeZ9l/HOJkUX/MI7bCmv
> eKrtQQium4cJfXx2fGGdR2O21g+oQML/V0UTHkq78qMr3Z9Bqj3wRRyVGc5ch7kb
> 3P8aYjlJG8qU6OtDiwvRNIMvAQzovetJu18L+F/X3lgxqsvrEjk1Ugdach0smmkl
> TcM9U21la7oOAxhFg4J5W6aatdDx6kKU6s4K68kpcvpQ0MJjQ6bhjKASzf/VFuTl
> 6vLXUiN+qHc/2yBmuxgMluRfAzFEsLWZ3Y6r3bQC4f3mfmUZxt7CJ8D+rYqSQKUW
> TDHABRVdUDJxbypUaA8q
> =VhSN
> -----END PGP SIGNATURE-----