OSNews

It depends on the clustering technology. If you use a MOSIX based system (which is what I would recomend, if you are not into scientific number crunching alone), you can load-balance seperate tasks quite nicely across a even heterogenous cluster (I have installed one for my sister who is not really a number cruncher, for parallising cd ripping to ogg files or render analog-to-digital videos from her camcorder at a process level. Very effective, even though the cluster employs several older machines). Mosix clusters are also known to work with scientific/mathematical software like (IIRC, please correct me if I'm wrong) octave or R.

I guess you could call me a professional number cruncher (computational physicist) and we use MPI (and sometimes even PVM) to parallise our integral equation solvers at code level (e.g. the programm has to implement the parallising structures), mostly due to the fact, that the equations kernels would never fit into a single machines memory.

If you programm in C/Fortran/C++ or Java, then playing with MPI to parallise existing code can be interesting and entertaining

EDIT: Some pointers for usage scenarios with openMosix, perhaps this gives you a better estimation of what general purpose applications can benefit from clusters.

[1] http://howto.x-tend.be/openMosix-HOWTO/x212.html
[2] http://povmosix.sourceforge.net/ [PovMosix = Povray + OpenMosix]
[3] http://www.imsc.res.in/~parapp/IMSCTalk.pdf

I have currently no access to my link collection, perhaps this three urls help a little

Edited 2007-01-29 18:13