Thanks guys. In response to 'Of course "personal computers" is now servers' which was stated by Douglas, If using servers is now the norm, well I'm still using PC towers :) This of course, Joe, is the reason I don't have to worry about Microsoft's policies on using XP on servers, since I am not. Does it matter if the nodes don't communicate? It seems like most communication between nodes would be a bottleneck. For example, if I'm doing A sum function (Like the ones used in pi calculation) which has two parts, instead of having the nodes communicate for each "part" of the sum, adding them as they went, one would distribute the calculations, have each node do a certain number of the parts, then add them at the end. Is that still parallel cluster-ing?
-- Sure... Parallel computation can be done at all manner of scales from very coarse (divvy up the work into independent chunks, farm it out, wait for them all to be done) to very fine: think of a finite element model where each node does one node of a grid, and then sends the results to each neighbor in the grid, so you have to communicate on every step. I suppose one could conceive of a Beowulf cluster using sneakernet of floppy disks as the communications pipe among nodes. In general, folks for low performance use Ethernet,BSD sockets, etc. >From a programming standpoint there's an MPI version for Windows (I assume it >works on XP.. it worked on NT 4.0 when last I fooled with it). MPI uses >whatever communications fabric you tell it do, but I doubt anyone has written >a sneakernet fabric.
_______________________________________________ Beowulf mailing list, [email protected] sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
