Measure elapsed time the right way
Depending on the low level Operating System (OS) settings, Ruby's
clock_gettime Linux functions from
According to the of
gettimeofday it returns a struct with the number of seconds and the system time zone. Ruby VM can then calculate and it returns a
Time object with these informations. This is often indicated as wall time in Linux documentation.
The time returned by
gettimeofday()is affected by discontinuous jumps in the system time (e.g., if the system administrator manually changes the system time).
The lack of perfect accuracy is due to CPU physical conditions like temperature, air pressure, and even magnetic fields.
So when the OS tries to set a new system time, it doesn't guarantee that the new value will be in the future. If a CPU has a clock that is "too fast", the OS can decide to reset the time of a few seconds backward. Another reason why wall clocks are flawed is because some CPUs can't manage leap seconds.
Posix systems have solved this problem by introducing a monotonic clock. It's conceptually similar to a timer that starts with an event and it isn't affected by time floating problems. Each time you request the time to the monotonic clock, it returns the time since that event. On Mac OS, this event is the system boot. Alongside with monotonic, there are several clock types: realtime, monotonic raw, virtual just to name a few. Each of them solves a different problem.
Since Ruby 2.1+ (MRI 2.1+ and JRuby 220.127.116.11+) there is a new method that allows to access to the current values of all these clocks:
This is named after the Linux function (still from
starting = Process.clock_gettime(Process::CLOCK_MONOTONIC) # time consuming operation... ending = Process.clock_gettime(Process::CLOCK_MONOTONIC) elapsed = ending - starting elapsed # => 9.183449000120163 seconds