Compare Linux CPU frequency governors

Create a governor configuration script

The governor experiment keeps the hardware minimum and maximum frequency limits unchanged. Only the governor changes between runs.

Create set-governor.sh:

    

        
        
cat > set-governor.sh <<'EOF'
#!/usr/bin/env bash
set -eu

governor=${1:?Usage: $0 GOVERNOR}

for policy in /sys/devices/system/cpu/cpufreq/policy*; do
    if ! grep -qw "$governor" "$policy/scaling_available_governors"; then
        echo "$governor is not available for $(basename "$policy")" >&2
        exit 1
    fi
done

for policy in /sys/devices/system/cpu/cpufreq/policy*; do
    hardware_max=$(cat "$policy/cpuinfo_max_freq")
    hardware_min=$(cat "$policy/cpuinfo_min_freq")

    echo "$hardware_max" | sudo tee "$policy/scaling_max_freq" > /dev/null
    echo "$hardware_min" | sudo tee "$policy/scaling_min_freq" > /dev/null
    echo "$governor" | sudo tee "$policy/scaling_governor" > /dev/null
done

echo "Configured all CPUFreq policies with the $governor governor"
EOF

chmod +x set-governor.sh

    

The script restores the full hardware frequency range before applying the governor. This prevents a frequency cap from an earlier experiment affecting the result.

Run the powersave test

Configure the powersave governor and verify that all policies changed:

    

        
        
sudo ./set-governor.sh powersave
grep -h . /sys/devices/system/cpu/cpufreq/policy*/scaling_governor | sort -u

    

The expected output is:

    

        
        powersave

        
    

Allow the system to settle, then run OpenSSL:

    

        
        
sleep 60
sudo ./run-openssl.sh governor-powersave 90

    

Run the schedutil test

Configure schedutil and run the same workload:

    

        
        
sudo ./set-governor.sh schedutil
sleep 60
sudo ./run-openssl.sh governor-schedutil 90

    

This run repeats the baseline configuration. Keeping it in the governor sequence makes it easier to compare runs collected close together under similar room and system conditions.

Run the performance test

Configure performance and run the workload:

    

        
        
sudo ./set-governor.sh performance
sleep 60
sudo ./run-openssl.sh governor-performance 90

    

Inspect the frequency traces

Print the minimum, average, and maximum observed frequency for each run:

    

        
        
for csv in results/governor-*/telemetry.csv; do
    awk -F, '
        NR == 2 { min = max = $3 }
        NR > 1 {
            total += $3
            count++
            if ($3 < min) min = $3
            if ($3 > max) max = $3
        }
        END {
            printf "%s min=%.0f avg=%.0f max=%.0f kHz\n", FILENAME, min, total / count, max
        }
    ' "$csv"
done

    

The performance governor should request the highest permitted performance. The powersave governor should request the lowest permitted performance. The schedutil governor adjusts the request according to scheduler utilization.

Check the measured frequency instead of assuming each governor behaved as expected. Firmware, platform controls, and thermal limits can affect the actual frequency.

Restore the default configuration

Return to the original settings after the governor tests:

    

        
        
./restore-cpufreq.sh

    

What you’ve accomplished and what’s next

You’ve now collected equivalent workload and telemetry data for three governors while preserving the full frequency range.

Next, you’ll keep schedutil fixed and change only the maximum permitted frequency.