Skip to content

Monitoring and Performance

System monitoring tools for CPU, memory, disk I/O, and network. This entry covers real-time monitors, snapshot utilities, resource limits, and control groups.

Memory

free -h                     # human-readable memory usage
free -m                     # in megabytes
cat /proc/meminfo           # detailed kernel memory info
vmstat 1 5                  # virtual memory stats every 1s, 5 times

Output columns: total, used, free, shared, buff/cache, available.

Disk I/O

iostat                      # I/O stats for devices
iostat -x 1                 # extended stats, refresh every 1s
iotop                       # per-process I/O (like top for disk)

Network

netstat -tulpn              # listening ports and services
ss -tulpn                   # same, newer/faster tool
netstat -an                 # all connections
iftop                       # real-time network usage by host
nload                       # bandwidth monitor

System Load and Uptime

uptime                      # uptime, users, load average
cat /proc/loadavg           # load average values

Load average = processes waiting for CPU + running. Three values: 1-min, 5-min, 15-min. - < CPU cores: healthy - = cores: at capacity - > cores: overloaded

Resource Limits

ulimit -a                   # show all limits
ulimit -n 4096              # set max open files

Control Groups (cgroups)

Kernel feature for limiting and isolating resource usage per process group.

Capabilities: - CPU pinning to specific cores - CPU time/weight limits - Memory limits (hard cap) - I/O limits (absolute or weighted) - Resource accounting and auditing

cat /proc/PID/cgroup        # view process cgroup
ls /sys/fs/cgroup/          # cgroup v2 hierarchy

Log-Based Monitoring

tail -f /var/log/syslog         # follow system log
tail -f /var/log/auth.log       # follow auth log
journalctl -f                   # follow systemd journal
journalctl -u nginx             # specific service
journalctl --since "1 hour ago"
journalctl -p err               # errors only

Gotchas

  • top load average includes D-state (disk wait) processes - high load does not always mean CPU bottleneck
  • free "used" includes buffers/cache - look at "available" for actual free memory
  • ss is preferred over netstat on modern systems (faster, more information)
  • iotop requires root and the kernel CONFIG_TASK_IO_ACCOUNTING option
  • High wa (I/O wait) in top indicates disk bottleneck, not CPU issue

See Also