Slashdot Mirror
Coolest Cluster Ever
sw155kn1f3 writes "Scientists at Los Alamos National Laboratory built a cheap (less than $1k per unit) 294-unit Beowulf claster dedicated to run astrophysics calculations. According to their website it's 85th fastest computer in the world. Seems cool and promising as it made with cheap components and off the shelf hardware."
It would be possible for a group of people, not necessarily a small group, but not necessarily huge, either to repeat this. 100 people, each with $3,000, could do it. The group would need to find some space to house the thing, and would probably have to do it in a climate where it could be relatively naturally cooled, which definitely rules out Phoenix. The computer would then be one of the fastest machines in the world.
Granted, I don't know what the hell they'd do with the computer, but it would be kind of cool to be on the list.
IBM had PL/1, with syntax worse than JOSS,
And everywhere the language went, it was a total loss...
For $294,000US, this has to impress you.
All of the top 10 fastest computers in the world are multi-multi-million dollar machines. This is a breakthrough because it represents another milestone in bringing supercomputing accessible.
$264k one day, $100k the next. I sincerely hope that soon small-to-medium enterprises can own supercomputers. With all the low budget physics stuff going on at Universities around the world, cheap supercomputing can only be a good thing.
What makes a man want to be a mouse? (Python's Flying Circus)
Take a look at this room A 1000-Pentium Beowulf-Style Cluster Computer half way down the page.
Analytic & algebraic topology of locally Euclidean meterization of infinitely differentiable Riemmanian manifold
Last modified: November 15, 2002
:P
We are the last to know about it
"Some fight for law. Some fight for justice. What will you fight for? One day, you will see."
The P4 has issues when you can't keep the pipeline full and with some stalling on branch misprediction. These PCs will be running simulations, meaning brutal amounts of linear algebra for the most part, most likely only a thread or two of computations per machine. You're certainly not going to get much performance out of the Crusoe.
Not to mention national security issues. LANL has had many problems with security in the last few years.
My other first post is car post.
At the semiconductor company where I work, I'm finding most of the EDA tools are being ported from expensive Sun Ultrasparcs to P4/Athlon-based commodity platforms with Linux OSes. And guess what? The processor clock speed has a direct correlation with performance compared to the slow-poke Ultrasparc 3s. You can reach a memory limit for some operations, but tools like Magma and our internal tools that are ported are running at least twice as fast per processor. Particularly with hierarchical designs, the only time the Sun Servers become necessary is for all the back end physical verification, parasitic extraction and signal integrity analysis, where less users are interactively spending their time anyway versus the floorplan/place/route anyway. So, whether I go out and buy an E4500 with 6 processors and 20GB of memory and use LSF, or I buy a dual Xeon 2.4GHz with 4GB of ECC and a Seagate HD, I'm getting a hell of a lot more mileage out of the dual Xeon and a huge cost benefit too for 10% of the entry cost of the Sun.
Sorry, but commodity PC hardware really does have a place in real computational work on the design of multimillion-gate standard-cell ASICs like the ones going into the latest Nvidia and ATI cards. The Suns are, for now, necessary, but it won't be long until commodity hardware usurps its place for a fraction of their overpriced niche monopoly in EDA tools.
I really like the picture on the sites frontpage.
i can imagine the small size of the Shuttles being an advantage, not to mention the "coolness" factor looking at it. (i assume the "cool" in the intro refers to emotion and not teperature!!)
But getting computation done cheap is no longer the challenge. It's getting the data from one node to the other. They still need "custom" expensive equipment for this.
I see they use 3com gigabit ethernet. having this 300+ gigabit switch capability is not "cheap".
Until one can buy this kind of networking equipment for really cheap, we shouldn't mention things like "of the shelve Beowulf super computer in the top 100".
Why are other peoples sig's always more witty ???
Disclaimer: I have never built a cluster before.
What is the advantage of having a hard drive in each node. Can't you boot each node off of a networked image and load the OS and whatever "work" into memory.
thanks
Journal
Has anyone ever written a Slashdot culture archive? Slashdot has a pretty rich and entertaining culture, and I hate to see things like Natalie Portman, Beowulf, and goats.cx...well, maybe not goats.cx....vanish forgotten into the mists of time. I've tremendously enjoyed cultural archives of USENET, where various trends or customs were explained, with links to example text.
I'd love to read something like this, if anyone ever gets around to setting up a website to archive these.
It's very difficult to identify trends (like, say, what the meaning of hot grits is) long after the fact -- you're looking at hundreds of thousands of old tech posts. But if someone is thoughtful enough to make a note that this is happening...well, five or ten years from now, it could be quite a fun to read little work.
May we never see th
Why didnt these guys post a HOWTO?
.. but it would be cool if they gave a little more HOWTO stuff in there.
Using Linux and all, yeah we appreciate they posted at least SOME info
Especially how the Foundry stuff is connected, it didnt seem clear to me.
Also, what kind of software do they ran to manage the cluster environment.
Enquiring minds want to know.
What's the bandwidth of that trunk? Also, what's the capaity of the connections between each 16-port card and the backplane?
Just curious... suppose all the units on a 16-port card have 1Gbps each, but only 8Gb total to the backplane. Then the backplane, in turn, has only 8Gb to the other switch. These are just made up numbers, but how would beowulf handle it? Can it group jobs requiring higher communication throughput onto the nodes which are closer to eah other? Does it have to be told the topology, or does it figure it out?
Sounds like they have 10 ports @ 1Gbit each free, so get about 10Gbit between the switches. When we (position 180 on the latest top500) were investigating thin-tree connections, we thought that we might be able to effectively run one job on a third of the nodes, another on the second third, and miscellaneous jobs on the 3rd (since we have two or 3 researchers who like to use lots of resources, and people like me who only need a single proc at a time). So you just partition the nodes and only allow your mpi jobs to sit on one group at a time.
Then there are things like openmosix which deal with topology automatically somehow. They will try to calculate the speed of the interconnect and the various nodes' procs (if heterogenous), and work out the distribution in the most efficient way. I am going to try to convince my sysadmin to try out openmosix on the lesser-used nodes of the cluster, because there is a feature of it that I think one of us might like - the combination of memory of the different nodes in one big contigous space, but right now we are busy cleaning up after the upgrade.