I like the idea of this mailing list. It sounds like you've got the
exact idea of what I'm interested in doing as well. Architecture-wise
I do have a few machines on which I could get away with such a
"distributed distributed compute" scheme. That's really what would be
cool. A reniced idle-eater job that requets jobs that are themselves
reniced idle-eaters. Perhaps even some job control. The CONDOR pointer
seemed interesting, but you have to relink your applications (not
really a big deal) In reality, I suppose this isn't any different from
any other distributed resource management scheme. You'd just want a
really light-weight one, with some minimal job control (like the ability
to terminate its children at will), designed with idle-eating in mind.
-Erik Kline
ISNTA