Infrastructure & Operations › Linux & Servers

Process Priority (nice)

Telling the scheduler which processes matter more.

Also known as: nice, renice, process priority

Linux shares the CPU between runnable processes using a scheduler. Process priority lets you influence who gets the CPU first. The mechanism is the nice value: an integer from -20 (highest priority, least “nice” to others) to 19 (lowest priority). Normal processes start at 0. Lower nice means more CPU.

nice -n 10 ./backup.sh        # start a process at lower priority
renice -n 5 -p 1234           # change a running process
nice                           # show the current niceness

Use it to keep bulk work from competing with latency-sensitive services: run a backup, a batch job or a re-index at nice 10 so interactive traffic wins when the machine is busy. Only root can lower the nice value (raise priority) below 0.

The classic mistakes:

• Thinking nice controls I/O. It only affects CPU scheduling. A backup is usually disk-heavy, not CPU-heavy, so nice barely helps. For disk bandwidth there’s a separate mechanism (ionice), and even that is best-effort depending on the scheduler and storage.
• Trying to raise priority without privilege. Lowering nice below 0 requires root; a normal user can only increase niceness (reduce priority).
• Setting it and forgetting. Nice is a nudge, not a guarantee. A high-priority process still waits if the machine is saturated; it doesn’t reserve capacity.
• Using it to fix a deeper problem. If a service needs priority, the real issue may be that too much unrelated work runs on the same box, or that it needs more capacity.

When not to bother: on a lightly loaded system, nice has little visible effect. It matters when a machine is oversubscribed and you want to bias the scheduler. Watch the load average and per-process tools to see whether the pressure is CPU at all — if it’s memory or disk, priority on the CPU won’t help.