How to get time (milisecond) of a python IO execution

Question

I am trying to test measure some IO execution in milliseconds , but bit confuse about best method achive that under windows 7. I am using following code but not sure if its best or correct way since i have two different results, which one should i take as results and which is best way. Please advice

code exec results 
100000000 loops, best of 3: 0.00925 usec per loop
3.827947176762156

run code
 import timeit ,time
 import datetime
 import os

 def main():    
 os.path.getsize("c:/video-2011-09-09-09-32-29.mp4")  
 if __name__ == '__main__':    
 t = timeit.Timer('main()')    
 print (t.timeit(1))

Answer 1

I'm not familiar with how Windows works, but I wouldn't be surprised if it caches directory information. (Unix systems certainly would.) You probably aren't really doing much actual I/O to get the size of a file after the first run. Also, you probably want to subtract the time it takes to execute a no-op function:

 def noop(): pass

to eliminate the overhead of calling your main function.

Answer 2

Measuring any performance can be tricky, and I/O in particular. Windows will cache the recently used file information, at a few levels. So if you do the same thing repeatedly, you'll get an unnaturally low average
time.

if you really need to measure how long it takes to get a file size, you would need to write your own, very controlled, test code. For example, there is an APi call that tells Windows to flush its disk caching logic. However, it doesn't necessarily flush the buffers that the drive itself keeps. Worse, in order to run your program, you have to do some disk I/O, and if that happens to use some of the same sectors as the thing you're timing, you're biasing the results.

If you have multiple partitions, you could arrange that the file in question is the only thing your system uses on that partition, to minimize the likelihood that something else has primed the pump. But realize also that once you're measuring the real thing, speed of exeution will vary enormously depending on the speed of the drive, the layout of the directories, and the fragmentation of the disk.