Re: how to set priority for idle process ?

Harald Koenig (root@tat.physik.uni-tuebingen.de)
Sat, 12 Apr 1997 01:54:40 +0200


On Apr 11, Greg Alexander wrote:

> > what's the best way to set the process priority (2.0.x kernel)
> > for a process which should *not* compete (or as less as possible)
> > with user processes running with "nice 19" ?
>
> Could you do "nice 20"?

nice 20 is the same as nice 19 (check ps output).

> Regardless, everything should compete equally if on
> the same niceness level. if the user doesn't want competitors to win the
> timeslice battle, they shouldn't nice themselves down that much.

I want to have 3 possible classes of applications/processes running:

1) interactice work
2) background comptations & batch jobs
3) idle task

and I'd like that the higher priority get *every* CPU time needed and
they shouldn't need to compete with lower priorities at all (some kind of
fixed priority scheduling).

running 3 jobs with nice 0, 10 and 19 give the following distribution
of CPU time:

PID USER PRI NI SIZE RES SHRD STAT %CPU %MEM TIME COMMAND
480 root 17 0 816 56 28 R 61.6 0.1 0:37 idle
479 root 18 10 816 56 28 R N 31.9 0.1 0:19 idle
478 root 20 19 816 56 28 R N 5.0 0.1 0:03 idle

so the interactive job looses ~40% cpu time due to background jobs.
only running background and idle jon (nice 10 and 19) gives

PID USER PRI NI SIZE RES SHRD STAT %CPU %MEM TIME COMMAND
483 root 18 10 816 56 28 R N 85.9 0.1 0:52 idle
482 root 20 19 816 56 28 R N 12.5 0.1 0:08 idle

so the idle task still uses 12% CPU while it shouldn't as there are
more important jobs around...

Harald

--
All SCSI disks will from now on                     ___       _____
be required to send an email notice                0--,|    /OOOOOOO\
24 hours prior to complete hardware failure!      <_/  /  /OOOOOOOOOOO\
                                                    \  \/OOOOOOOOOOOOOOO\
                                                      \ OOOOOOOOOOOOOOOOO|//
Harald Koenig,                                         \/\/\/\/\/\/\/\/\/
Inst.f.Theoret.Astrophysik                              //  /     \\  \
koenig@tat.physik.uni-tuebingen.de                     ^^^^^       ^^^^^