AcademySoftwareFoundation
determineIdleCores
determineIdleCores reduces idle cores by using the load average.
https://github.com/AcademySoftwareFoundation/OpenCue/blob/c22fe12721bed24b814184ea78c5e447b3b6a477/cuebot/src/main/java/com/imageworks/spcue/dispatcher/DispatchSupportService.java#L609-L615
However, the actual system load of the host doesn't look like so busy (sar -u 1 or other metrics don't show much things to worry about). Maybe using load average might not be a good idea on Linux?
First post ever to Github, but this is more in my wheelhouse so felt I would not embarrass myself. It sort of depends on whether you are looking for the state of the host or the state of process'. Load is more like a time averaged queue of things using a cpu. CPU activity is more akin to a snapshot of activity much shorter duration. I think load is what you are looking for here. It removes things like IO waits, is a measurement over time of overall demands to use a cpu rather than a ratio of execution/waits and is a simpler number to get "right".
A better explanation here that deep dives the farther you go! https://www.linuxjournal.com/article/9001
My inclination is to agree with @ultiplyr -- load seems like what we want. Is there a proposal to use some alternate metric? I'm sure we'd be happy to discuss.
It's also something we could add as a config setting to allow users to customize.
