iostat cheatsheet


A cheatsheet of iostat in Linux. Mostly learned from Interpreting iostat Output

Background:

  • iostat reports at physics device/sector level (i.e. beneath cache and IO scheduling)

Overall utilization:

  • avg-cpu / %iowait: how busy is the CPU, or “amount of computation waiting for IO”
  • device / %util: how busy is the device

Columns (listed per device):

  • rrqm/s wrqm/s: r/w requests merged per second
    • Block IO subsystem may merge physically adjacent requests
  • r/s w/s: number of (possibly merged) r/w requests
  • rsec/s wsec/s: number of sectors read/written
    • iostat can also be set to use units rkB/s wkB/s rMB/s wMB/s
  • avgrq-sz: average size of request, in #sectors
    • larger size indicates sequential IO, smaller indicates random IO
  • avgqu-sz: average length of request queue, i.e. average #request waiting to be served
  • await: average (wait + serve) time for a request, in ms
    • r_await / w_await: the same time for r/w requests
  • svctm: average serve time for a request, in ms

A monitor command with iostat:

# -x: extended report
# -t: print time for report
# -z: omit inactive devices
# 1: report every 1 second

$ iostat -xtz 1

<!– TODO: iostat in BSD >