iMac Beowulf Cluster Comes to Life
AmigaAvenger writes "Finally a good use for all those old IMacs that many organizations have laying around collecting dust. We have set up a 5 node (4+1 controller) iMac beowulf cluster, which is appropriately named Skittles, and is running PPC Yellowdog Linux, with MPICH 1.2.6 cluster message passing software."
They should have lined them up in a circle like in the old iMac posters!
Taking guns away from the 99% gives the 1% 100% of the power.
I can understand the geek factor, even the teaching / experimenting factor. But, if anybody is in the same situation, please donate those macs to a charity. There are milions of kids in this world that have no access to IT, and could benefit from this kind of equipment. Even if its net value is below 100$, you can bet some kid could play or learn on it.
I'll do the stupid thing first and then you shy people follow...
..."Why?"
Not why do it. But why post about it.
This may be interesting if the cluster was created to serve some purpose. but if the purpose was merely for the admins to learn about how to set up a Beowulf cluster, well, that exercise is probably performed at least once a day somewhere in the world.
Hardly seems to warrant a post on one's own Web site, let alone a link to Slashdot, IMHO.
"Skittles"...Cute name, though. Wait til Mars, Inc. hears about it...
---anactofgod---
"Equal opportunity swindling - *that* is the true test of a sustainable democracy."
Hasn't it been pretty much shown that you can cluster pretty much anything given a network connecting the elements together? What seems to be lost is that the point of clustering is to achieve a certain amount of performance for a lower cost using multiple cheap elements than the alternative higher prices single high performance system. The whole point is lost when the amount of time and effort that goes into the project costs more than the equivalent single out-of-the-box system. In this case, you're seeing on the order of tens of megaflops with what is likely to be a rather high node-node latency. It costs in the high hundreds to low thousands of dollars now to get on the order of 100 MFLOPS to a few GFLOPS in a single box without the latency penalty incurred by message passing programs on the cluster.
So, in addition to beating a dead horse yet again, this project is actually a waste of money if cycles are what you're looking for. And last time I checked, that's pretty much all clusters are for.