performance measurement questions

[Eric S. Johansson]

I've been playing with the profiler and have been rather successful in
making my code run faster. So for my best success has been cutting a
three second CPU time run down to about 0.45 seconds.

Unfortunately, the same application seems to take about 8 to 12 seconds
real-time. I know that some of that time is lost to file IO and
external sub processes but I don't know how much.

Is there any way to change the profiler to measure elapsed time and not
CPU time?

 

[Roy Smith]

I've been playing with the profiler and have been rather successful in
making my code run faster. So for my best success has been cutting a
three second CPU time run down to about 0.45 seconds.

Unfortunately, the same application seems to take about 8 to 12 seconds
real-time. I know that some of that time is lost to file IO and
external sub processes but I don't know how much.

Is there any way to change the profiler to measure elapsed time and not
CPU time?

I don't know about the profiler, but there's a couple of things you
could do to try and figure out what's going one (all of these assume a
unix-like environment):

First, use the "time" command to run your program, like this:

Roy-Smiths-Computer:unit$ time ./TargetUnit.py
........
----------------------------------------------------------------------
Ran 8 tests in 1.064s

OK

real 0m1.340s
user 0m0.960s
sys 0m0.100s

That should give you a rough idea what's going on.

If you have some ideas where the time might be spent internally, you can
use the os.times() function to gather time snapshots around areas you
suspect of being slow.

 

[Eric S. Johansson]

I don't know about the profiler, but there's a couple of things you
could do to try and figure out what's going one (all of these assume a
unix-like environment):

First, use the "time" command to run your program, like this:

unfortunately, the programs in question are CGI's and servers. I need
something networks inside the program.

If you have some ideas where the time might be spent internally, you can
use the os.times() function to gather time snapshots around areas you
suspect of being slow.

when it comes to program performance measurement, guessing usually gives
you a wrong answer since my request for profiling that measures elapsed
time.

unfortunately, I may need to do as you suggest in place time functions
throughout the code.

---eric