I have put together a �Cluster that addresses some of these concerns. Photogs of it can be seen here...
http://www.extreme-linux.com/pix As you can probably see from the pix, I have a lid that opens up so that you can maintain the nodes. I reversed the power connections on one side to bury the cables in the belly of the box; I have palm switches that go in there as well. It is arranged so that the MoBos function much like blades; I have (since the pix were taken) put an ATX umbilical together so that when you pull the MoBo out it springs loose. Putting it back in is a little trickier. I eschewed latency concerns in favor of bandwidth ones and implemented a channel-bonding scheme much like the one used at http://ilab.usc.edu/beo. Marty Connor wrote a wonderful PXE HOWTO at www.ltsp.org/documentation/pxe.howto.html that pretty much worked right out of the box for me... I need to make some more design changes before it is ready for primetime and then I am going to loan it to Jon Proulx (of this list and elsewhere) to evaluate for a while.... I agree with many of Jorge's suggestions (I ran his page through the URL translator at worldlingo.com) and for raw heat-spewing, number-crunching, beowulf-thrashing performance, AMD CPUs are the way to go... ...now if only we could get the data to and from the AMD MoBos faster! *********************************** Kurt Keville Debian Beowulf Users Group (DeBUG) http://www.extreme-linux.com Cluster On! ***********************************
