Solaris 10, MPI, C++ volunteer needed

[UncleAl]

Uncle Al has parallelized code in C++ that uses MPI,

 

LAM/MPI Parallel Computing

 

This puppy needs 50K+ CPU-hrs to run. Sun Microsystems may donate the time in its 5000-Opteron grid. It requires skills in C++/MPI/Solaris 10 to compile the code (compiler and time free) and oversee the app as it runs (call it 30 CPUs for 70 days; about 3 MB data total output; retrieve data weekly). One must tread lightly in the Sun grid: learn-as-you-go is inappropriate.

 

Does anybody skilled in the art want to play? Reply here, by private message, or

 

organiker(coupling symbol)lycos(period)com

 

Thank you!

[alexander]

I would love to play, but i would be the learn-as-you-go type *shrug*

 

oh you dont actually compile the mpi code on the cluster, you can compile it and test it on any gcc compiler with mpi lib linking... you could then use a beowulf cluster, amazon or even the mit cluster to run it (and i can setup a beowulf cluster at school, well if i get permission, which i would :hihi: prolly be a good 150 to 200 Go of total world domination power, for 2-3 days max if i had to guess...) You just need good threading, and stuff like that

 

what are you calculating, uncle?

[UncleAl]

The CPU-hrs matter. Hardware is testable, any OS not Windows. If a cluster is sitting around unloved, romance is in the air! Beowulf is good.

 

It calculates the handedness of chiral crystal lattice mass distribution with increasing radius. Run time grows as the unit cell population and (radius)^2. This supports two challenges of General Relativity versus teleparallel gravitation. GR has an unexamined loophole involving left and right hands. Try to get academics to think outside their own skulls! Numbers help.

 

http://www.mazepath.com/uncleal/qzsparse.png

Looks good, but it's not. Needs denser sampling

http://www.mazepath.com/uncleal/qzdense.png

250,000 CPU-hrs. 9 atoms/unit cell. Quartz is complete.

It's not noise, It's real variation. We went sparse at the end.

http://www.mazepath.com/uncleal/bzhdense.png

Benzil, 78 atoms/unit cell. 1300 CPU-hrs AMD Athon FX-55

Another 700 CPU-hrs in progress

 

Total run time in any radius interval can be calculated from three calibration points run in one CPU, 36 min 4 sec in the FX-55. No run time surprises. The code is debugged and documented. It ran quartz in a 16-Opteron cluster for 40 days. The serial version ran 6 weekends in a 168-Pentium cluster (to locate defective cores). C++ source code is locally compiled for the cluster that runs it, ditto MPI linking as you said.

 

Quartz is 90 days in $2 million equipment - difficult to arrange. Benzil is one week in two calorimeters - a cosmology group is starting to sniff. Benzil beyond 65,000 A radius and 107 trillion atoms in a single CPU - even in a very fast CPU - is not practical. We'd like 140,000 A radius and a quadrillion atoms, out to log(radius) = 5.15 on the graph. We suffer "but what happens at larger radii?" Physicists want numbers.

 

Does it sound like fun? (Get permission. Weekends and holidays often lack users.)

[Turtle]

...

Quartz is 90 days in $2 million equipment - difficult to arrange. Benzil is one week in two calorimeters - a cosmology group is starting to sniff. Benzil beyond 65,000 A radius and 107 trillion atoms in a single CPU - even in a very fast CPU - is not practical. We'd like 140,000 A radius and a quadrillion atoms, out to log(radius) = 5.15 on the graph. We suffer "but what happens at larger radii?" Physicists want numbers.

 

Does it sound like fun? (Get permission. Weekends and holidays often lack users.)

 

Can't help with the CPU; all the wrong stuff. :( Nonetheless, it does sound fun. :doh: Is someone going to run your experiment?

 

That's all I got. ;)

[UncleAl]

This past Christmas an analytical firm at 45 degrees latitude volunteed its idle thermal analysis lab. Nice folk! Alas, though OEM spec was 0.1% precison, tested precision was no better than 3.5%. One DSC's accuracy was off by 40%. Accuracy disappears in a difference measurement. Precision is not negotiable.

 

Calorimetric Equivalence Principle Test

 

We need an adventuresome manufacturer, an academic with elegant standards, or a commercial lab that hasn't used its hardware to near death.

 

Calorimetric Equivalence Principle Test

 

GR can't be quantized, string theory is a disaster; supersymmetry predictions farted, the axion telescope CAST is a no-show, dark matter experiments are uniformly nulls, the best spin experiment shot craps as did the best EP probes,

 

http://www.npl.washington.edu/eotwash/publications/pdf/prl97-021603.pdf

http://www.npl.washington.edu/eotwash/publications/pdf/schlamminger08.pdf

 

The only new idea arising from orthodox theory is mine - do local left and right hands vacuum free fall identically? They'll come around even if they must be bludgeoned with extended calculations. What is the worst it can do - fail? That is the gold standard of performance.

 

First, more numbers to be crunched. You've got to sing to them in their own language.