nice and renice: priorities

Sharing one CPU

A CPU core runs one thing at a time; the kernel's scheduler creates the illusion of parallel work by handing out short slices of time. When several processes want the CPU, each one's nice value tilts the share. The scale goes from -20 (greediest, highest priority) to 19 (nicest, lowest priority), with 0 as the default. The name is literal: a process with a high nice value is being *nice* to the others. Note what nice does *not* do: on an idle machine a process at 19 still runs at full speed. Priority only matters when there is competition.

~% yes > /dev/null &
~% nice -n 19 yes > /dev/null &
~% ps -o pid,stat,args | grep '[y]es'
  924 R    yes
  926 RN   yes
~% top -b -n 1 | head -6
CPU:  93% usr   6% sys   0% nic   0% idle ...
  PID  PPID USER     STAT   VSZ %VSZ %CPU COMMAND
  924   923 root     R     1340   1%  80% yes
~% killall yes

Two copies of yes fight for the single core of the lab VM. The normal one takes about 80 % and the niced one is starved so badly it does not even reach the top of the list. ps marks it with N in STAT (low priority), the flag you met in lesson 52; a negative nice value shows as < instead.

nice: start with a priority

~% nice
0
~% nice -n 5 nice
5
~% nice sleep 600 &
~% nice -n 15 sleep 600 &
~% nice -n 25 nice
19
~% nice -n -5 sleep 600 &

Alone, nice prints the current niceness. In front of a command it adds to it: nice CMD adds 10, nice -n N CMD adds N, and the result is clamped to the -20…19 range. Children inherit the value, so nice -n 10 sh gives you a whole shell whose every command is niced. Only root may go down: a normal user can make their own processes nicer but never greedier, and cannot undo a raise. In the lab you are root, so nice -n -5 works; on a shared server it would fail with *Permission denied*.

renice: change a running process

~% sleep 900 &
[1] 1062
~% renice 15 1062
~% awk '{print $19}' /proc/1062/stat
15
~% renice -n 2 1062
~% awk '{print $19}' /proc/1062/stat
17
~% renice 5 -p 1062 1063

renice changes processes that are already running. In BusyBox, renice VALUE PID sets the nice value to exactly VALUE, while renice -n STEP PID adds STEP to the current one. -p (the default) says the IDs are PIDs; -g and -u apply the change to a whole process group or to every process of a user. Other implementations disagree about -n, so on any unfamiliar system check the result afterwards. BusyBox ps has no column for the number itself, only the N and < flags; the exact value is field 19 of /proc/PID/stat, which is where the awk above reads it (lesson 60 explores /proc).