Establish an OpenSSL performance and power baseline

Use SHA-256 as the workload

The OpenSSL speed command measures cryptographic throughput without downloading a dataset or building an application. SHA-256 provides a CPU-intensive server workload with a direct result in bytes processed per second.

Use a fixed 16384 byte buffer so every run performs the same operation. Use one OpenSSL worker per online CPU to load the full processor.

Create the workload runner

Create run-openssl.sh in ~/perf-per-watt:

    

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

label=${1:?Usage: $0 LABEL [SECONDS]}
duration=${2:-90}
result_dir="results/$label"

mkdir -p "$result_dir"

policy=/sys/devices/system/cpu/cpufreq/policy0
cat "$policy/scaling_driver" > "$result_dir/scaling_driver.txt"
cat "$policy/scaling_governor" > "$result_dir/scaling_governor.txt"
cat "$policy/scaling_min_freq" > "$result_dir/scaling_min_freq.txt"
cat "$policy/scaling_max_freq" > "$result_dir/scaling_max_freq.txt"
openssl version -a > "$result_dir/openssl-version.txt"

echo "Warming up all CPUs for 30 seconds"
openssl speed \
    -elapsed \
    -seconds 30 \
    -multi "$(nproc)" \
    -bytes 16384 \
    -evp sha256 \
    > /dev/null 2>&1

echo "Collecting telemetry for $duration seconds"
./collect-telemetry.sh "$result_dir/telemetry.csv" &
logger_pid=$!

cleanup() {
    kill "$logger_pid" 2>/dev/null || true
    wait "$logger_pid" 2>/dev/null || true
}

trap cleanup EXIT INT TERM

openssl speed \
    -elapsed \
    -seconds "$duration" \
    -multi "$(nproc)" \
    -bytes 16384 \
    -evp sha256 \
    2>&1 | tee "$result_dir/openssl-output.txt"

echo "Results saved in $result_dir"
EOF

chmod +x run-openssl.sh

    

The runner performs a 30-second warm-up before collecting results. The warm-up loads OpenSSL code and data, raises CPU activity, and reduces the effect of starting from an idle state.

Confirm the baseline settings

The initial run uses the current CPUFreq settings. Confirm that the first policy still uses the schedutil governor and full frequency range:

    

        
        
grep . /sys/devices/system/cpu/cpufreq/policy0/{scaling_driver,scaling_governor,scaling_min_freq,scaling_max_freq}

    

On the Thelio Astra example, the baseline is:

    

        
        scaling_driver:cppc_cpufreq
scaling_governor:schedutil
scaling_min_freq:1000000
scaling_max_freq:2200000

        
    

Run the baseline workload

Allow the system to remain idle for 60 seconds before the run:

    

        
        
sleep 60
sudo ./run-openssl.sh baseline-schedutil 90

    

The final OpenSSL table contains the aggregate SHA-256 throughput for the 16384 byte buffer size. Keep the complete output file because OpenSSL output formatting can vary between versions.

List the files captured for the run:

    

        
        
find results/baseline-schedutil -maxdepth 1 -type f -printf '%f\n' | sort

    

The expected files are:

    

        
        openssl-output.txt
openssl-version.txt
scaling_driver.txt
scaling_governor.txt
scaling_max_freq.txt
scaling_min_freq.txt
telemetry.csv

        
    

Check the telemetry duration

The logger contains about one sample per second. Count the samples:

    

        
        
awk 'END { print NR - 1, "telemetry samples" }' results/baseline-schedutil/telemetry.csv

    

A 90-second run produces approximately 90 samples. A small difference is normal because process startup and shutdown don’t align exactly with the sampling interval.

What you’ve accomplished and what’s next

You’ve now collected a baseline OpenSSL throughput result and synchronized telemetry using the default CPUFreq configuration.

Next, you’ll change only the governor and compare its effect.