Understanding CPU Usage Computation

CPU usage essentially measures the time a processor spends executing a process. In Linux systems, this execution time is tracked in units called jiffies. By calculating jiffies * HZ, we can determine the CPU time consumed by a process. Dividing this value by the total CPU time yields the CPU usage percentage: jiffies * HZ / total_time.

Contrasting ps and top

Both ps and top are widely used for monitoring CPU usage, yet they employ fundamentally different methodologies.

We can demonstrate this difference using a simple Go program:

package main

import (
	"bytes"
	"fmt"
	"strconv"
	"sync"
	"time"
)

var testData = []byte(`testdata`)

func testBuffer(idx int) {
	m := make(map[string]*bytes.Buffer)
	for i := 0; i < 100; i++ {
		buf, ok := m[strconv.Itoa(i)]
		if !ok {
			buf = new(bytes.Buffer)
		}
		for j := 0; j < 1024; j++ {
			buf.Write(testData)
		}
		m[strconv.Itoa(i)] = buf
	}
	fmt.Println("done, ", idx)
	wg.Done()
}

var wg sync.WaitGroup

func main() {
	for i := 0; i < 10; i++ {
		wg.Add(1)
		go testBuffer(i)
	}
	wg.Wait()
	fmt.Println("sleeping")
	time.Sleep(time.Hour)
}

Observing the Discrepancy

Running this program and checking its CPU usage with top and ps reveals notable differences:

Using top -n 1:

PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND
39753 infini    20   0 14.663g 0.014t   1200 S 611.1 22.2   0:23.53 test-cpu

Using ps aux:

USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
infini   39881  767 39.1 26505284 25791892 pts/16 Sl+ 07:04   0:38 ./test-cpu

After the testBuffer completes, top reports near-zero CPU usage while ps still shows high usage:

USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
infini   39881 82.3 42.4 28638148 27953532 pts/16 Sl+ 07:04   0:40 ./test-cpu

Explaining the Difference

The discrepancy arises from their execution methods. top operates over a duration, while ps provides an instantaneous snapshot. This affects their calculation methods:

ps calculates average CPU usage over the process's lifetime:

(total_cpu_time / total_process_uptime)

This means even after CPU usage drops, ps will report gradually decreasing values as the process continues running.

top, however, computes usage over a measured interval:

(current_cpu_time - last_cpu_time) / iteration_duration

Continuous CPU Usage Monitoring

Monitoring systems typically consist of data collection and visualization components. While ps is suitable for collecting instantaneous values, it only provides average usage rather than real-time metrics.

For accurate real-time monitoring, we need to collect raw CPU time data and calculate usage based on sampling intervals:

delta(cpu_time) / delta(timestamp)

In Linux, both tools obtain data from /proc/[PID]/stat:

• utime: CPU time spent in user code (jiffies)
• stime: CPU time spent in kernel code (jiffies)
• cutime: User code time including chlidren (jiffies)
• cstime: Kernel code time including children (jiffies)

Using derivative functions to compare these values against timestamps provides accurate real-time CPU usage metrics.