sleep, time and measuring a command

sleep: doing nothing, precisely

You have used sleep all through this module as a harmless process to look at. Its real job is to wait: a script that retries a download every ten seconds, a loop that polls until a file appears, a pause between two steps that must not overlap. sleep N waits N seconds and costs no CPU while it does; the kernel simply does not schedule it until the timer fires. BusyBox sleep accepts fractions (0.5), the suffixes s, m, h and d, and several arguments that are added together, so sleep 1m 30s waits ninety seconds. For really short pauses there is also usleep, which takes microseconds.

~% sleep 0.5
~% sleep 1m 30s &
~% ps -o args | grep '[s]leep 1m'
sleep 1m 30s
~% until [ -f /tmp/ready ]; do sleep 1; done; echo ready

time: how long did that take?

~% time sleep 2
real	0m 2.00s
user	0m 0.00s
sys	0m 0.00s
~% time wc -l /etc/passwd
9 /etc/passwd
real	0m 0.00s
user	0m 0.00s
sys	0m 0.00s

time CMD runs the command and, when it exits, reports three numbers. real is wall-clock time, what a stopwatch would show. user is CPU time spent running the program's own code, and sys is CPU time the kernel spent on its behalf (reading files, allocating memory). Comparing them is the useful part. sleep 2 shows real 2 and almost zero CPU: it was waiting, not working. A CPU-bound job shows real close to user plus sys. When real is much larger than user plus sys, the command spent its time waiting: for the disk, the network, a lock, or for the CPU itself because something with a better nice value was running (lesson 57).

Here, time is a program

In bash, time is a shell keyword that can time anything, pipelines and loops included. In this VM's ash it is the BusyBox applet /usr/bin/time (type time says so), an ordinary program that runs one other program. That has consequences. It cannot time builtins: time cd / fails with *can't execute 'cd'*. In time sort big | uniq -c, only sort is timed. The fix for both is to hand the whole thing to a shell: time sh -c 'sort big | uniq -c > /dev/null'. And the report is written to standard error, so time CMD > out.txt still prints it on the screen, while time CMD 2> t.txt captures it (along with any error messages of CMD).

~% time -p sleep 1
real 1.01
user 0.00
sys 0.00
~% time -p -o /tmp/t.txt sleep 1
~% cat /tmp/t.txt
real 1.01
user 0.00
sys 0.00
~% time -f '%e s' sleep 1
1.00 s
~% time cd /
time: can't execute 'cd': No such file or directory

Measuring by hand

~% start=$(date +%s); sleep 3; end=$(date +%s)
~% echo "took $((end - start)) s"
took 3 s

Inside a script you often want to time a whole section, not one command. Saving date +%s (seconds since 1970, lesson 68) before and after and subtracting is crude, whole seconds only, but works anywhere and is perfectly good for jobs that take minutes.