Python IS slow ! [was] Re: Python too slow for real world
Markus Kohler
markus_kohler at hp.com
Tue May 4 06:17:18 EDT 1999
>>>>> "Guido" == Guido van Rossum <guido at eric.cnri.reston.va.us> writes:
Guido> Markus Kohler <markus_kohler at hp.com> writes:
>> >>>>> "Marko" == Marko Schulz
>> <4mschulz at informatik.uni-hamburg.de> writes:
>>
Marko> Markus Kohler <markus_kohler at hp.com> wrote:
>> >> Python-1.5.2c1 .... 0.52 seconds >> >> Squeak .... 0.13
>> seconds.
>>
Marko> I wouldn't put too much meaning in this number. You can't
Marko> say how big the startup costs are for a python vs. squeak.
>> I'm not really interested in the startup costs. One can make
>> them near zero for squeak, by compiling the image into an
>> executable. These are the times without startup costs.
Guido> Markus, I think Marko was asking how you measured the
Guido> times. If you used something like "time python bench.py"
Guido> you were measuring Python's (considerable) start-up time,
Guido> and you are misrepresenting Python's speed. On the other
Guido> hand, if you used Python's time.clock(): t1 = time.clock()
Guido> bench() t2 = time.clock() you're fine. Without this
Guido> information your numbers are meaningless (at least to
Guido> skeptics :-).
Ok, I admit that I also tested with time python bench.py.
But the result is in this case almost the same. because I repeated the
test 100 times. That means the test runs almost one minute, startup time
becomes unimportant for such a small test.
Remeasuring with time gave me the same result.
By the way the call I used is tak(18, 12, 6). I forgot to mention that in
my first post.
Anyway, what I just want to say is that there is room for improvement and that
it's worth the effort.
I also measured the function call speed of the language 'ruby' that
R. Beherends mentioned here. It runs the function call speed test
almost at the same speed as python.
Markus
--
Markus Kohler mailto:markus_kohler at hp.com
More information about the Python-list
mailing list