I think this is such good news, that even those on sage-devel will not mind too much! Building packages in parallel seems to be working very well. That said, wtih 128 hardware threads, still only a small fraction of 't2' is being used effectively. I've found with multi-threaded coded, around 500 threads seems optimal on 't2' - certainly a lot more than the number of cores (16) or hardware threads (128).

Clint, the main ATLAS developer, sent an email the other day to William in which he said the latest ATLAS would build in under an hour on 't2', so hopefully 'ATLAS' wont be a bottleneck for much longer.

Now at least 't2' should build Sage quicker than the 10-year old SPARC I bought from eBay for a US equivalent of under $50.

Dave

-------- Original Message --------
Subject: building on t2 in 4 hours
Date: Wed, 9 Jun 2010 16:02:49 -0700 (PDT)
From: John H Palmieri <jhpalmier...@gmail.com>
Reply-To: sage-sola...@googlegroups.com
To: sage-solaris <sage-sola...@googlegroups.com>

I just built Sage on t2.math.washington.edu in four hours, with only
one cheat:

 - The cheat: I used an installation of atlas from a previous Sage
build.

This saves 10-12 hours.  With this cheat, I have in the past been able
to build Sage in 17 hours.  To cut it down to 4 hours: install the
patches at #8306 (parallel inter/intra-spkg building). The result
(after fixing one small mistake in SAGE_ROOT/spkg/standard/deps):

real    243m24.846s
user    833m14.300s
sys     129m48.406s

...

Sage build/upgrade complete!

(The "real" line is the key one.  Since it was building parallel, the
"user" time is longer than the actual elapsed time.)

I'm happy to give #8306 a positive review, but it depends on some
other tickets -- #8731, #9185, #9186, and #9187 -- so please review
those.  Several of them involve very small changes to spkg-install
files, but someone more familiar than I am with makefiles and related
issues should review them.

--
John

--
To post to this group, send an email to sage-devel@googlegroups.com
To unsubscribe from this group, send an email to 
sage-devel+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org

Reply via email to