OK, I was confusing OpenMosix with Beowolf. Beowolf clusters are clusters of PCs running Linux to do massive number crunching with specialized parallel processor comiplers and such.

Well, early on in the Beowolf project they started out with the "embarssingly parallel supercomputer" model. It went like this. If you had to run the same ellaborate calculation on several different numbers, you simply ran the calculation program on several different computers each for a different number and then pooled all the results when each program run terminated.

Can this be done on an OpenMosix cluster? As I understand it, I could link all my PCs on my LAN in such a cluster but start and monitor several processes from one PC, right?

For instance, say I want to calculate a billion digits of pi as accurately as possible in the shortest amount of time possible. I could evaluate the area under the curve y=4/(1+x^2) from 0 to 1 (=4arctan(1)-4arctan(0)=pi). Instead of evaluating this on one PC, I could assign 1/25th of this area to one PC, another 1/25th to another and so on. Could I start my area program from the same PC for all processes and eventaully the processes that can't be accomodated by the resources of my PC will migrate to other PCs in the cluster as needed? Can I then get the results of all 25 processes on my one PC when they are done to add them all together and get pi? I would use a reimann sum program for this with many subdivisions, hence the ellaborate calculation!

A better program may be to approximate the entire area from 0 to 1 with a power series and just evaluate as many terms as you need for millionth place accuracy! OOO, this sounds good. So, if I need 1000 terms, just run 1000 processes shared over 25 PCs!

This would be a great and easy intro to the world of supercomputing for my high school computer science students!

TIA