9.2 Adjust Process Scheduling

9.2 Adjust Process Scheduling

Quick Answer: CPU scheduling preference uses nice values from -20 (favor this process) to 19 (run only when others do not need CPU). Start with nice -n N command. Change live processes with renice N -p PID. Confirm the NI column via ps or top. Root can apply aggressive (negative) nice values; unprivileged users are limited.

Where this fits among “operate running systems”

After you can find and kill hot processes (Section 9.1), the next official skill is to adjust process scheduling. On RHEL this means the classic UNIX nice mechanism that biases the kernel's default fair-share CPU scheduler. On RHEL 10 (kernel 6.12) that scheduler is EEVDF (Earliest Eligible Virtual Deadline First), which replaced the older CFS (Completely Fair Scheduler) in upstream Linux 6.6. The nice/renice interface and the -20..19 range are unchanged by that switch. This is not real-time chrt policy work unless a task explicitly goes there (rare on EX200; master nice/renice first).

Typical tasks:

• “Start /usr/local/bin/batchjob with nice value 10.”
• “Set the nice value of process report (or PID …) to 5.”
• “Ensure the long backup runs at lower priority than interactive work.”

You are changing how hard the process competes for CPU, not stopping it and not configuring tuned profiles (Section 9.3).

Nice value scale

Mnemonic: A higher nice number is nicer to everyone else—your process yields. A negative nice is aggressive and usually root-only.

Related fields you may see:

Do not try to memorize every PRI mapping. EX200 wording almost always says nice value or “priority” in the nice sense—set NI correctly.

Starting processes with nice

nice command
# default nice when using nice without -n is typically +10 on many systems
nice -n 10 /usr/local/bin/batchjob
nice -n 19 tar czf /backup/home.tgz /home
sudo nice -n -5 /usr/local/bin/urgent-job

Notes:

• nice applies to the command you start and is inherited by children unless they renice themselves.
• Without -n, traditional behavior starts at nice 10 (check man nice on the exam host if unsure)—always pass -n explicitly on the exam so the value is unambiguous.
• Background example:

nice -n 15 longtask.sh > /var/tmp/longtask.log 2>&1 &

Changing running processes with renice

renice 10 -p 2234
renice -n 10 -p 2234          # GNU form also common
sudo renice -5 -p 2234        # boost priority (negative nice)
renice 5 -u student           # all of user's processes (powerful—use carefully)
renice 5 -g 1000              # by process group / GID forms per man page

Prefer PID form on EX200 unless the task says “all processes for user X.”

# Discover then renice
pgrep -a report
# 55201 report
sudo renice 5 -p 55201
ps -o pid,user,ni,comm -p 55201

Permission rules (critical)

If renice fails with “Permission denied” or “failed to set priority,” escalate with sudo when the task requires a privileged value or another user’s PID.

Verification

ps -o pid,user,ni,pri,comm -p PID
ps -eo pid,ni,comm --sort=ni | head
top -p PID
# In top, NI column must match the required value

Example success line:

  PID USER      NI COMMAND
55201 root       5 report

Also:

cat /proc/PID/stat   # advanced; nice is among the fields—ps is clearer under time pressure

nice vs kill vs services vs tuned

Do not renice systemd (PID 1) as a “fix.” Do not assume renice persists across process restart: a new process gets default nice unless the launcher uses nice again, a service unit sets Nice=, or a wrapper script applies it.

systemd Nice= (awareness)

Some service units include:

Nice=10

in the unit file. Editing units is a services/deploy skill. For “adjust process scheduling” tasks that name a running process, renice is the direct tool. If a task requires a service to always run niced after reboot, you may need unit drop-ins—read the task; do not invent unit edits when a simple renice satisfies “set nice of this process to N now.”

Real-time scheduling (brief, do not over-study)

chrt -p PID
chrt -f 10 command    # SCHED_FIFO example—dangerous if misused

Real-time priorities can starve the system. EX200 study points emphasize nice-style adjustment. Only use chrt if a scenario explicitly demands it. Prefer nice/renice for nearly all practice.

Exam workflows

Workflow 1 — Launch with explicit nice

Task: Run /usr/local/bin/crcscan at nice 15.

nice -n 15 /usr/local/bin/crcscan &
# or foreground if the task expects a long interactive run
pgrep -a crcscan
ps -o pid,ni,comm -C crcscan

Workflow 2 — Fix an already running job

Task: Process datamunch must have nice value 10.

pgrep -a datamunch
sudo renice 10 -p $(pgrep -n datamunch)   # -n newest match; confirm only one PID first!
# Safer:
pgrep datamunch
sudo renice 10 -p 44120
ps -o pid,ni,comm -p 44120

Avoid blind $(pgrep ...) if multiple PIDs match—renice accepts multiple -p or run a loop after listing.

for pid in $(pgrep datamunch); do sudo renice 10 -p "$pid"; done
ps -o pid,ni,comm -C datamunch

Workflow 3 — Lower priority without killing

A compile job starves the GUI/SSH interactivity. Soften it:

pgrep -a make
sudo renice 19 -p 3001

The job keeps running; interactive sessions feel better under load.

Workflow 4 — Privileged boost

Task: Set nice of PID 2200 to -10.

sudo renice -10 -p 2200
ps -o pid,ni,comm -p 2200

Without root this fails—use sudo on the exam VM.

Interaction with multi-CPU and load

Nice does not pin a process to a CPU (that is taskset). Nice does not limit memory (systemd-run -p MemoryMax= or cgroups). Nice only tilts CPU time sharing when runnable processes compete. On an idle system, even nice 19 may run at full speed because nothing else wants the CPU—that is normal.

Common traps

1. Inverting the scale — thinking 19 is “highest priority.” It is lowest.
2. Omitting -n and assuming the default is 0 when using the nice wrapper (default is often +10).
3. Renicing once then restarting the app—new PID loses the setting.
4. Using kill -STOP by mistake while trying to “slow” a process—STOP freezes it entirely.
5. Confusing NI with %CPU — high %CPU can still have NI 0; renice changes preference, not an instant %CPU cap.
6. renice without verifying — always re-check ps -o ni.

Persistence notes

• renice affects the current process until it exits or is reniced again.
• Reboot clears all user-space PIDs; nothing about a one-time renice survives unless the job is started again with nice or a unit/cron wrapper sets it.
• For exam tasks phrased as “set the nice value of this running process,” immediate renice + verification is enough.
• For “always run this batch at nice 10,” combine scheduling tools (cron/systemd timer) with nice in the command line—covered more fully under task scheduling objectives.

Section checkpoint

You should explain the -20..19 nice scale, start jobs with nice -n, change live jobs with renice, know when root is required, verify with ps/top NI, and avoid confusing scheduling bias with kill or tuned. That meets the EX200 “adjust process scheduling” study point.