Slashdot Mirror
Simulating the Universe with a zBox
An anonymous reader writes "Scientists at the University of Zurich predict that our galaxy is filled with a quadrillion clouds of dark matter with the mass of the Earth and size of the
solar system. The results in this weeks journal Nature, also covered in Astronomy magazine, were made using a six month calculation on hundreds of processors of a self-built supercomputer, the zBox. This novel machine is a high density cube of processors cooled by a central airflow system. I like the initial back of an envelope design. Apparently, one of these ghostly dark matter haloes passes through the solar system every few thousand years leaving a trail of high energy gamma ray photons."
Looks like MS will need to come up with a new name for the Xbox 3.
At the center of the box is a small piece of fairy cake
zBox has been slashdotted, vat do ve do now? Ve relax and eat some cheese and chocolate, jawohl!
[Picture of loads a wires in what looks like a thousand desktops interlinked]
Our in-house designed (Joachim Stadel & Ben Moore 2003), massively parallel supercomputer for running our cosmological N-body simulations. This machine consists of 288 AMD Athlon-MP 2200+ (1.8 GHz) CPUs within a few cubic meters. Under load it produces about 45 kW of heat, about equivalent to 45 electric hair dryers operating continuously! This amount of heat, combined with the extremely high density necessitated a new design for efficient cooling. The 144 nodes (2 CPUs per node) are connected using an SCI fast interconnect supplied by Dolphin in a 12x12 2-dimensional torus. The layout of the machine is ring-like, thereby allowing very short "ribbon" cables to be used between the nodes. This fast interconnect network attains a peak bisection bandwidth of 96 Gbits/sec, with a node-node write/read latency as low as 1.5/3.5 microseconds. Additionally the zBox has 11.5 TBytes of disk (80 GBytes/node) and 3 Gbits/s I/O bandwidth to a frontend server with 7 TB of RAID-5 storage. This is among the fastest parallel computers in the world! At "first light" it ranked in the top 100, but the technology advances quickly. (see top500, June 2003: Rank 144) (see top500, November 2003: Rank 276)
We greatly acknowledge the aid of the Physics Mechanical Workshop at the University of Zurich for: 1) turning the "napkin-sketch" into a proper CAD/CAM design of the machine; 2) providing numerous suggestions which improved the detailed design; 3) providing a gigantic room for the construction of the boards; 4) and, well, building the thing! We thank the companies of Dolphin (dolphinics.com) for supplying the high speed network and COBOLT Netservices for supplying the majority of parts. We would like to especially thank the individuals: Doug Potter and Simen Timian Thoresen for their great help in setting up the linux kernel and root file system, getting netbooting to work correctly, and resolving several operating system related problems. Finally we thank all who helped in the construction of the zBox (assembly of boards, etc), Tracy Ewen, Juerg Diemand, Chiara Mastropietro, Tobias Kaufman
...you just slashdotted Switzerland. Who's next, tough guy? Andorra?
Chip H.
You can read the entire paper in PDF or PS at astro-ph, a web site which collects preprints in the physical sciences. See
http://xxx.lanl.gov/abs/astro-ph/0501589
I read the paper quickly. The authors have to come up with a model which has virtually no observable consequences (otherwise, we would have seen this source of matter by now), but which can also be tested experimentally in the not-too-distant-future (or else it wouldn't be science). They predict that some of the cosmic-ray shower telescopes may be able to detect the little cloudlets of dark matter. We'll see.
Michael Richmond "This is the heart that broke my finger."
mwrsps@rit.edu http://stupendous.rit.edu
So one of these dark matter clouds may pass through the solar system every few thousand years? Have they taken the next step and hypothesized that such an event could account for major climate changes? Like the event that killed off the dinosaurs?
http://krone.physik.unizh.ch/~stadel/zBox/story.ht ml
The 3D temperature monitor is really cool.
Maybe they should have use the zbox to host their site =)
i k.unizh.ch/~stadel/zBox/
http://rufus.hackish.org/~rufus/mirror/krone.phys
All it keeps saying is 42...42...42...42...
You can legislate morally you can't legislate morality
The problem in question is the number of distinguishable bodies. With weather you would have to go down to the single molecule in the air, to get a quite good prediction. In fact current weather models use cubes of air where the conditions are considered constant (same temperature, same pressure, same direction of air flow in the same cube) and take them as distinguishable bodies. Those models are a compromise between the sheer number of necessary elements, the number crunching limits of current calculation hardware and the difference between the used model and the reality.
With stellar bodies it's much more easy. The number of stellar bodies you need for a prediction is much smaller, the bodies themself can be considered almost constant for the whole calculation etc.pp. With the number crunching capacity of today's weather prediction centers you can simulate whole galaxies (if you consider stars constant, which they mainly are for about 10mio to 10bio years, depending on their mass). With the differences between your model and the measured reality you can spot elements you didn't simulate yet and add them to your model. The swiss team now was simulating clouds of about the mass of the earth and the size of the solar system and found that those added to the stellar simulation made a quite good fit to the measured data.
but then I read in the article:
"and had sufficient forced air through the heat exchangers to transport the heat from a small car out of this small room."
Suprising.
There's nothing Intelligent about Intelligent Design.
The reason I'm interested is that a non-neutral charge distribution would tend to attract the outer part of the galaxy towards the centre more than would be expected from gravity alone, which is (simplistically) the evidence for dark matter / energy.
"All 288 CPUs shipped by AMD worked perfectly and none needed to be replaced" My 500Mhz AMD works perfectly... as long as I use reliable software (Linux) not that other product - what's it called again... XPee?
I actually thought myself a few months ago about putting a group of 4 HDs and 4 mobos on a large aluminium plate, placing in a wide, flat enclosure and feeding air in at the center and out via 4 peripheral ducts to build a 4-way unit that could sit under a set of office desks arranged roughly in a square. The benefit is that the hardware takes up zero usable desk space, is well protected from physical damage, and the under-desk air flow results in low noise. For high density offices (e.g. call centers) with all power and network connections feeding in to the center of the desk clusters, this could be a very efficient arrangement. It's nice to know I was beaten to it by a Swiss supercomputer.
Panurge has posted for the last time. Thanks for the positive moderations.
If the computer ran for 6 months straight using 1.8GHz processors, couldn't they have waited several months and utilized newer CPUs running at double the speed, halving the computation time?
Regarding their design, I'm somewhat surprised they used an individual power supply for each board. It seems there would be more efficient and smaller power systems available that could power multiple boards at once. It looks like a quarter of the volume of the computer is comprised of power supplies. Plus all that extra heat is thrown into the mix too.
Dan East
Better known as 318230.
My guesses are: 1) Cost. A commercial 1U dual processor pizza box is actually very expensive for the computing power, compared to the do-it-yourself method. Of course, you're mostly paying for support, overhead, the brand name, and a sturdy, cool-looking case. 2) Cost. Commerical racks can be pretty pricey, too. 3) I/O speed. The zBox is wired up for good inter-CPU throughput, whereas you lose significant speed with the typical ethernet patchboard scheme you find in a commerical rackspace. Of these (3) is probably the most important.
If you mod me down, I shall become more powerful than you can possibly imagine.
a self-built supercomputer
I thought we where years away from having to defend ourselves against the machines...
Thats alot of porn!
http://www.livejournal.com/users/cixel
If you were trying to predict rain in the next six months it would be a lot easier than predicting it with any real useful accuracy.
It's the difference between saying it does rain, and when it will. On this scale they are just explaining a phenomena that can happen every so often, in a stellar sense. I'm guessing this eases the difficulty of computation from what would be necessary to predict the number of years before the next occurrence.
Of blankness, I know nothing.
I'm still waiting for a simulation engine that models the subatomic and atomic particles' behaviors.
Basically, I want a program that simulates chemical reactions. If I have a bunch of molecules mixed together, and I add another mixture, what will happen, on the atomic level?
We have SPICE for electrical circuits. Why not something for chemical reactions?
--TheOrangeSquid Is it any wonder things seem so awry? We swim in a sea of confusion and don't have to think to survive
If I remember correctly, chemical reactions happen at the atomic level, not subatomic.o l/chemtool.html
Here is a GPL program, there are plenty of others (commercial and FOSS):
http://ruby.chemie.uni-freiburg.de/~martin/chemto
"The problem in question is the number of distinguishable bodies. With weather you would have to go down to the single molecule in the air, to get a quite good prediction"
This is utter nonsense.
Weather models are fluid models, not particle models like the N-body simulation described here. They are quite different, and require different computational approaches. Both are numerically intensive, however.
Well, you'd have to have a capture/drawing tool like Chemtool, and then something that could approximate polarity, electrical charge distribution, and bond length/strength. (Those involve things like electron orbitals, hence the subatomic.) Next, you'd have to have something that handles movement of fluids and gases with respect to the temperature, pressure, etc (gas laws and the partial diff. eqns. whose exact solution is one of the Clay instutite Millenium problems). Then, you'd have to have something that will predict what happens, probabilistically, when two or more molecules interact. These interactions would have to modeled in terms of molecular collisions, so that things like titration, stirring, etc, would be accurate.
Finally, you'd have something which would prepare an "answer" to each problem by waiting for a reasonable amount of precipitate to settle, or measuring pH, or simulating a gas chromatograph of the contents of the beaker.
Other helpful things would be crystallization and such. I would think that if you could simulate the physical laws and properties at a sufficiently low level, most things would arise automatically, but IANAC.
--TheOrangeSquid Is it any wonder things seem so awry? We swim in a sea of confusion and don't have to think to survive