> I think it makes no sense in todays configurations and workloads. It 
   > can
   > be turned on via sysctl, but it should not be on by default.
   
   I agree, disable it.  Really, it should never have been added in the 
   first place.  Judging by some scheduler fixes that went in after 
   autonicetime was added, I can only assume that it was a misguided hack 
   to mask over some of those scheduler bugs.

the behaviour comes from 4BSD and used to be hard-coded to happen
to all non-root procs that used >10mins of CPU and had nice=0.  in
the old days we used to renice daemons to 1 early to avoid them
being reniced to 4 at some later time...for a long time my kernels
#if 0'ed that code out.  later i started using the sysctl's after
they were added...


.mrg.