Excessive install/build time for Sage under Linux?

[DuaneKaufman]

Hi,

I downloaded the sage 2.9.3 tarball to my notebook (Debian, Centrino
1.5Ghz, 2Gb RAM), and started a build.

That was about 2 days ago, and it is _still_ grinding away at the
build (still seems to be progressing, plenty of memory free, no disk
thraashing)

Is this normal? Other than this, nothing else seems amiss with the
functioning of the notebook. Is there anything else I can check?

According to top, there is a _lot_ of time spent running programs like
xctfc, and other x<nnnn> variants

Thanks,
Duane.

[mabshoff]

On Jan 11, 2:37 pm, DuaneKaufman <duane.kauf...@gmail.com> wrote:
> Hi,

Hello Duane,

> I downloaded the sage 2.9.3 tarball to my notebook (Debian, Centrino
> 1.5Ghz, 2Gb RAM), and started a build.
>
> That was about 2 days ago, and it is _still_ grinding away at the
> build (still seems to be progressing, plenty of memory free, no disk
> thraashing)
>
> Is this normal? Other than this, nothing else seems amiss with the
> functioning of the notebook. Is there anything else I can check?
>

You have hit an ATLAS build bug, namely that ATLAS mis-detects Pentium
Ms as CoreDuos (see http://trac.sagemath.org/sage_trac/ticket/1740).
That has been fixed for the next, i.e. Sage 2.10, release.

To solve this kill the build, download
http://sage.math.washington.edu/home/mabshoff/release-cycles-2.10/alpha1/atlas-3.8.p7.spkg
and copy that file into spkg/standard. There should already be
atlas-3.8.p6.spkg in that directory. Then restart the build via make
from the base directory of Sage. If the build keeps running that long
you have hit another bug. You should also disable power management
while building Sage (you have a laptop after all). For instructions
see the question "QUESTION: Sage 2.9 and higher fails compiling ATLAS
on Linux. How can I fix this?" at http://wiki.sagemath.org/faq

> According to top, there is a _lot_ of time spent running programs like
> xctfc, and other x<nnnn> variants

Yep, those are the tuning routines, which in your case do a complete
tuning, which is very, very expensive. With a pretuned build ATLAS
takes about 16 minutes to build on a Opteron 248.

> Thanks,
> Duane.

Let me know if anything else odd happens or if you have any more
questions.

Cheers,

Michael

[DuaneKaufman]

Indeed, I finally got through the build with your suggestion, though
it took 1900 minutes. Is this normal?

Thanks for your help,
Duane

On Jan 11, 7:55 am, mabshoff <Michael.Absh...@fsmath.mathematik.uni-

dortmund.de> wrote:
> On Jan 11, 2:37 pm, DuaneKaufman <duane.kauf...@gmail.com> wrote:
>
> > Hi,
>
> Hello Duane,
>
> > I downloaded the sage 2.9.3 tarball to my notebook (Debian, Centrino
> > 1.5Ghz, 2Gb RAM), and started a build.
>
> > That was about 2 days ago, and it is _still_ grinding away at the
> > build (still seems to be progressing, plenty of memory free, no disk
> > thraashing)
>
> > Is this normal? Other than this, nothing else seems amiss with the
> > functioning of the notebook. Is there anything else I can check?
>
> You have hit an ATLAS build bug, namely that ATLAS mis-detects Pentium

> Ms as CoreDuos (seehttp://trac.sagemath.org/sage_trac/ticket/1740).

> That has been fixed for the next, i.e. Sage 2.10, release.
>

> To solve this kill the build, downloadhttp://sage.math.washington.edu/home/mabshoff/release-cycles-2.10/alp...

> and copy that file into spkg/standard. There should already be
> atlas-3.8.p6.spkg in that directory. Then restart the build via make
> from the base directory of Sage. If the build keeps running that long
> you have hit another bug. You should also disable power management
> while building Sage (you have a laptop after all). For instructions
> see the question "QUESTION: Sage 2.9 and higher fails compiling ATLAS

> on Linux. How can I fix this?" athttp://wiki.sagemath.org/faq

[William Stein]

On Jan 14, 2008 6:39 AM, DuaneKaufman <duane....@gmail.com> wrote:
>
> Indeed, I finally got through the build with your suggestion, though
> it took 1900 minutes. Is this normal?

No, that is not normal. On a 1.8Ghz Opteron under Linux using
one processor, it takes 168 minutes (< 3 hours) real time to
completely build Sage-2.10.alpha2. Thus there must have been an issue
with ATLAS tuning still.

On a 2.6Ghz core 2 duo it takes 109 minutes real time to build sage-2.9.3.

-- William

--
William Stein
Associate Professor of Mathematics
University of Washington
http://wstein.org

[mabshoff]

On Jan 14, 3:44 pm, "William Stein" <wst...@gmail.com> wrote:

> On Jan 14, 2008 6:39 AM, DuaneKaufman <duane.kauf...@gmail.com> wrote:
>
>
>
> > Indeed, I finally got through the build with your suggestion, though
> > it took 1900 minutes. Is this normal?
>
> No, that is not normal. On a 1.8Ghz Opteron under Linux using
> one processor, it takes 168 minutes (< 3 hours) real time to
> completely build Sage-2.10.alpha2. Thus there must have been an issue
> with ATLAS tuning still.

Yep. The problem is that Pentium Ms do not have a pre-tuned profile,
but 1900 minutes is excessive. I posted Paul's fix to ATLAS'
sourceforge tracker, but I hadn't heard back yet. There are plans for
2.10 to let people use the system's ATLAS if they want to (similar to
Fortran), so in the future one can at least work around this issue
without the need to go into the build system and do that manually.

> On a 2.6Ghz core 2 duo it takes 109 minutes real time to build sage-2.9.3.
>
> -- William
>

Cheers,

Michael