Nice, renice, and CPU scheduling priorities · Linux

Linux has two scheduling-priority concepts that confuse newcomers. Nice value is the polite request: “give me less CPU if something more important needs it”. Real-time priority is the strict demand: “give me CPU before anything except equal or higher real-time tasks”. The two are different mechanisms with different use cases.

Nice value

The nice value is an integer from -20 (highest priority, most CPU) to +19 (lowest priority, least CPU). The default is 0.

Configuration changenice— nice -n N COMMAND runs COMMAND with nice value N. The default is 0. nice -n 5 reduces priority (gives way to other processes); nice -n -5 raises it (demands more CPU). Non-root users can only raise the nice value (lower priority); root can lower it.

$ nice -n 5 ./batch-job.sh; nice -n -10 ./critical-monitor.sh

Illustrative output

Configuration changerenice— renice -n N -p PID adjusts the nice value of a running process. renice -n N -u USER adjusts the nice value of every process owned by a user. The same permission rules apply: non-root users can only raise nice values for processes they own.

$ renice -n -5 -p 1234; renice -n 10 -u alice

Illustrative output

Read-only / Safetop with NI— top shows the NI (nice) column. PR is the kernel's mapped priority (20 + nice). %CPU shows actual usage. The idle % (id) is 87.5% — the system has spare CPU; this nginx is not under contention.

$ top -n 1 -b | head -10

top - 12:00:00 up 30 days,  3:45,  1 user,  load average: 4.20, 3.10, 2.50
Tasks: 234 total,   3 running
%Cpu(s): 12.5 us,  0.0 sy,  0.0 ni, 87.5 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
MiB Mem :  16000.0 total,   1234.5 free,   8901.2 used,   5864.3 buff/cache

PID USER   PR  NI   VIRT   RES   SHR  S  %CPU  %MEM     TIME+  COMMAND
12345 root   20   0  1234m  456m  123m  R  95.0   1.1   0:42.15  nginx
12346 www-d  25   5  1234m  456m  123m  S   2.0   1.1   0:01.20  nginx
...

Illustrative output

Real-time priority

Linux has two real-time scheduling classes: SCHED_FIFO (first-in, first-out) and SCHED_RR (round-robin). Real-time tasks have priority 1-99; higher is more urgent.

Configuration changechrt— chrt -f -p PRIORITY PID sets SCHED_FIFO with priority. chrt -r -p PRIORITY PID sets SCHED_RR. Real-time priorities 1-99 are higher than any non-real-time priority. Root only.

$ chrt -f -p 50 1234; chrt -r -p 10 1234

Illustrative output

CPU affinity

Configuration changetaskset— taskset -cp MASK PID sets the CPU affinity. MASK is a comma-separated list or range of CPU numbers. Use 0-3 to allow CPUs 0-3; use 0 to pin to CPU 0 only. Useful for cache locality (pin related processes to the same CPU) and for keeping noisy processes off specific cores.

$ taskset -cp 0-3 1234; taskset -cp 0 1234

pid 1234's current affinity mask: 0-3
pid 1234's current affinity mask: 0

Illustrative output

Configuration changetaskset + nice— Combine taskset and nice. The processes runs with elevated priority, but only on CPUs 0-3 — important on large hosts where pinning a process to a NUMA node affects memory access latency.

$ taskset -c 0-3 nice -n -5 ./process.sh

Illustrative output

When nice does not help

nice is a polite request to the scheduler. It does not:

Grant more CPU than is available.
Bypass cgroup limits (CPUQuota in systemd unit files).
Override real-time priorities.
Reduce I/O wait time (the process is still waiting for the disk).

If a process is CPU-bound and the host has spare CPU, lowering the nice value (raising priority) gives it more CPU. If the host is saturated, lowering the nice value of one process starves others — the overall throughput is bounded.

Production discipline

Practice	Why
Use nice for batch jobs.	Nightly ETL, backups, log rotation — they should yield to interactive load.
Never use real-time priority for production daemons.	A runaway real-time task can lock the host.
Use CPU affinity when NUMA locality matters.	On large hosts, memory access across NUMA nodes is 30-50% slower.
Never disable the CFS bandwidth control.	`CPUQuota` is the right way to throttle a service.

Nice value

Real-time priority

CPU affinity

When nice does not help

Production discipline

Knowledge check