[Python-Dev] performance delta with .py presence v.s. only pyc on python 2.7.x?

[John Smith]

My team has a python app that appears to perform 10+% better when the
python source is present v.s. when only the pyc's are distributed.  This is
contradictory to my understanding of how python leverages pyc's.

Scenario

1. pyc-only install sees mediocre performance. (pyc's are built using
compileall.py, then source .py's removed before packaging)
2. adding the .py files to the installed filesystem results in 10+%
performance improvement
3. verified no difference in mod times, no new .pyc's generated
4. remove the .py's and again see drop in performance


Env info: centos 6.5, x86_64, however python used is 32bit from the i686
repos (this is how we deal with some 32bit only c libraries that the app
must use)


Has anyone got an idea of what I might be running into?  The app itself is
a black box to me.  I was not able to isolate this using a simple piece of
python code that worked on a math problem for a discernible amount of time,
so it could be something to do with how the app is using python.  The app
was originally written for python 2.4...

Any suggestions on what I can try is appreciated.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/python-dev/attachments/20141007/252b869d/attachment.html>