Process priority and tuned profiles

The idea

Priority: nice values

When more processes want the CPU than there are cores, the scheduler decides who runs. You influence it with the nice value, from -20 (most favoured) to 19 (most polite). The default is 0. A nicer process (higher number) yields to others.

The rule to remember: a lower number means higher priority. Ordinary users can only make their own processes nicer; only root can lower a nice value or set a negative one.

Command	Does
nice -n 10 command	Start a command with nice value 10
sudo nice -n -5 command	Start with a higher priority (root only)
renice -n 10 -p PID	Change a running process
renice -n 5 -u alex	Change every process of a user
ps -o pid,ni,comm -p PID	See the value (NI column)
top then r	Renice interactively

Nice affects CPU share, not disk or network. It matters only when the CPU is contended; on an idle machine a nice-19 process still gets all the CPU.

Tuning: tuned profiles

Linux has hundreds of kernel settings (disk scheduler, CPU frequency policy, virtual memory, network buffers). tuned applies a coherent set of them, a profile, matched to the machine’s job. It runs as a service and applies the profile at boot.

Profile	Optimised for
balanced	General use, saves some power when idle
throughput-performance	Servers doing a lot of work: disk and network throughput
virtual-guest	A virtual machine (default when tuned detects one)
latency-performance	Low response time at the cost of power
network-latency, network-throughput	Network-heavy workloads
powersave	Laptops and low power

tuned-adm recommend inspects the machine (virtual or physical, type) and names the profile Red Hat suggests. That is the right answer for “set the recommended profile”.

Try it

ps -eo pid,ni,comm --sort=ni | head -4

    PID  NI COMMAND
     21 -20 kworker/R-kblockd
     22 -20 kworker/R-ata_sff
    731  -7 rtkit-daemon

Start and change

nice -n 15 sha256sum /dev/zero &
ps -o pid,ni,comm -C sha256sum

[1] 3011
    PID  NI COMMAND
   3011  15 sha256sum

renice -n 10 -p 3011
renice -n 0 -p 3011

3011 (process ID) old priority 15, new priority 10
renice: failed to set priority for 3011 (process ID): Permission denied

Raising the value (more polite) worked, lowering it back did not: a normal user cannot undo being nice. Root can: sudo renice -n 0 -p 3011.

Choose a tuned profile

sudo dnf install -y tuned
sudo systemctl enable --now tuned
tuned-adm active
tuned-adm recommend

Current active profile: virtual-guest
virtual-guest

tuned-adm list | head -5

Available profiles:
- balanced                    - General non-specialized tuned profile
- latency-performance         - Optimize for deterministic performance at the cost of increased power consumption
- powersave                   - Optimize for low power consumption
- throughput-performance      - Broadly applicable tuning that provides excellent performance across a variety of common server workloads

sudo tuned-adm profile throughput-performance
tuned-adm active
sudo tuned-adm verify
cat /etc/tuned/active_profile

Current active profile: throughput-performance
Verification succeeded, current system settings match the preset profile.
See tuned log file ('/var/log/tuned/tuned.log') for details.
throughput-performance

tuned-adm profile takes effect immediately and is saved in /etc/tuned/active_profile, so it is applied again at boot as long as the tuned service is enabled. If tuned is not enabled, the setting is lost on reboot, which is the usual trap in a “must persist” task.

Going further

A custom profile is a directory in /etc/tuned/profiles/NAME/ containing tuned.conf; it can start with [main] and include=balanced and override a few [sysctl] or [cpu] settings. tuned-adm profile_info NAME prints a profile’s contents, and tuned-adm profile A B merges two. tuned-adm verify is also a check for drift after someone changes a sysctl by hand. Priority has more levers than nice: chrt sets real-time scheduling policies, taskset pins a process to CPUs, and systemd units can set Nice= and CPUWeight= for services.