RHCSA Rapid Track

Process priority sets the importance of each process. Linux implements scheduling policies that define the rules by which processes are organized and prioritized to obtain CPU processing time. The various Linux scheduling policies might be designed to handle interactive application requests, or non-interactive batch application processing, or real-time application requirements. Real-time scheduling policies still use process priorities and queues. However, current, non-real-time (normal) scheduling policies use the Completely Fair Scheduler (CFS), which instead organizes processes that are awaiting CPU time into a binary search tree. This process priority uses the SCHED_NORMAL or SCHED_OTHER policies as the default scheduling policy.

Processes that run under the SCHED_NORMAL policy are assigned a static real-time priority of 0, to ensure that all system real-time processes have a higher priority than normal processes. However, this static priority value is not included when organizing normal process threads for CPU scheduling. Instead, the CFS scheduling algorithm arranges normal process threads into a time-weighted binary tree, where the first item has the lowest previously spent CPU time, and the last item has the most cumulative CPU time.

The order of the binary tree is additionally influenced by a user-modifiable, per-process nice value, which ranges from -20 (increased priority) to 19 (decreased priority), with a default of 0. Processes inherit their starting nice value from their parent process. All users can adjust the nice value to decrease priority, but only root can increase its priority.

A higher nice value indicates a decrease in the process priority from the default, or making the process nicer to other processes. A lower nice value indicates an increase in the process priority from the default, or making the process less nice to other processes.

Increasing the nice value lowers the thread's position, and decreasing the value raises the thread's position.

Privileged users can decrease the nice value of a process, to make a process less nice. A process is then repetitively placed higher in the binary tree, and therefore is scheduled more often. On a saturated system, the overall CPU time available to other processes is reduced.

Unprivileged users can only increase the nice value on their own processes, which makes their own processes nicer, and therefore lowers their placement in the binary tree. Unprivileged users cannot decrease their processes' nice values to raise their importance, nor can they adjust the nice values for another user's process.

Nice values map to a priority value, and both values are available for viewing in process listing commands. A nice value of -20 maps to a 0 priority in the top command. A nice value of 19 maps to a 39 priority in the top command.

Figure 6.3: Priorities and nice values as reported by the top command

In the preceding figure, the nice values are aligned with the priority values that are used by the top command. The top command displays the process priority in the PR column, and the nice value in the NI column. The top priority numbering scheme, which displays real-time process priorities as negative numbers, is a legacy of internal priority algorithms.

The following output is the summary and a partial process listing in the top command:

Tasks: 192 total,   1 running, 191 sleeping,   0 stopped,   0 zombie
%Cpu(s):  0.0 us,  1.6 sy,  0.0 ni, 96.9 id,  0.0 wa,  0.0 hi,  1.6 si,  0.0 st
MiB Mem :   5668.6 total,   4655.6 free,    470.1 used,    542.9 buff/cache
MiB Swap:      0.0 total,      0.0 free,      0.0 used.   4942.6 avail Mem

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
      1 root      20   0  172180  16232  10328 S   0.0   0.3   0:01.49 systemd
      2 root      20   0       0      0      0 S   0.0   0.0   0:00.01 kthreadd
      3 root       0 -20       0      0      0 I   0.0   0.0   0:00.00 rcu_gp
      4 root       0 -20       0      0      0 I   0.0   0.0   0:00.00 rcu_par_gp

The ps command displays process nice values, when using the default formatting options.

The following ps command lists all processes with their process ID, process name, nice value, and scheduling class. The processes are sorted in descending order by nice value. In the CLS scheduling class column, TS stands for time sharing, which is another name for the normal scheduling policies, including SCHED_NORMAL. Other CLS values, such as FF for first in first out and RR for round robin, indicate real-time processes. Real-time processes are not assigned nice values, as indicated by the dash (-) in the NI column. Advanced scheduling policies are taught in the Red Hat Performance Tuning: Linux in Physical, Virtual, and Cloud (RH442) course.

[user@host ~]$ ps axo pid,comm,nice,cls --sort=-nice
  PID COMMAND          NI CLS
   33 khugepaged       19  TS
   32 ksmd              5  TS
  814 rtkit-daemon      1  TS
    1 systemd           0  TS
    2 kthreadd          0  TS
    5 kworker/0:0-cgr   0  TS
    7 kworker/0:1-rcu   0  TS
    8 kworker/u4:0-ev   0  TS
   15 migration/0       -  FF
...output omitted...

You can change the nice value of an existing process with the renice command. This example uses the process ID from the previous example to change from the current nice value of 15 to a new nice value of 19.

[user@host ~]$ renice -n 19 2740
2740 (process ID) old priority 15, new priority 19

You can also use the top command to change the nice value on an existing process. From the top interactive interface, press the r key to access the renice command. Enter the process ID, and then enter the new nice value.