2.3-2.5 what improved?

[Robin Becker]

A large cgi based web Python-2.3 application needs to be speed improved.
experiments show the following under reasonable testing (these are 2 second
reportlab pdf productions)

1) 2.3 --2.5 improvement small 1-2%
2) cgi --fcgi improvement medium 10-12%

I sort of remember claims being made about 2.5 being 10% faster than 2.4/2.3 etc
etc. Can anyone say where the speedups were? Presumably we have a lot of old
cruft that could be improved in some way eg moving loops into comprehensions,
using iterator methods etc. Are those sort of things what we should look at?
--
Robin Becker

[robert]

Robin Becker wrote:

A large cgi based web Python-2.3 application needs to be speed improved.
experiments show the following under reasonable testing (these are 2
second reportlab pdf productions)

1) 2.3 --2.5 improvement small 1-2%
2) cgi --fcgi improvement medium 10-12%

I sort of remember claims being made about 2.5 being 10% faster than
2.4/2.3 etc etc. Can anyone say where the speedups were? Presumably we
have a lot of old cruft that could be improved in some way eg moving
loops into comprehensions, using iterator methods etc. Are those sort of
things what we should look at?

Python 2.5 became quite fat. For bare CGI the Python load/init
time eats all improvements. Smaller scripts even loose lot of speed.

I still like Python 2.3 for many other reasons for many
applications - especially for CGI's, on Windows, for deployable
apps, GUI's etc. because the fat coming with Python 2.4 is not
balanced by necessary goods - mostly just fancy things.
( I run even a list of patches and module copies/addaptations down
to 2.3 because of that :-) )

Real news come with Py3K
Robert

[bruno.desthuilliers]

Robin Becker a écrit :

A large cgi based web Python-2.3 application needs to be speed improved.
experiments show the following under reasonable testing (these are 2 second
reportlab pdf productions)

1) 2.3 --2.5 improvement small 1-2%
2) cgi --fcgi improvement medium 10-12%

I sort of remember claims being made about 2.5 being 10% faster than 2.4/2.3 etc
etc. Can anyone say where the speedups were?

AFAIK, most of the speedup comes from optimization of the builtin dict
type, which is the central
data structure in Python. But anyway, as Robert pointed out, using CGI
means lauching
a new Python process for each and every HTTP request.

Presumably we have a lot of old
cruft that could be improved in some way eg moving loops into comprehensions,
using iterator methods etc. Are those sort of things what we should look at?

Consider profiling your code before doing anything else - unless you're
planning on wasting your time.

[billie]

robert wrote

Robin Becker wrote:

A large cgi based web Python-2.3 application needs to be speed improved.
experiments show the following under reasonable testing (these are 2
second reportlab pdf productions)

1) 2.3 --2.5 improvement small 1-2%
2) cgi --fcgi improvement medium 10-12%

I sort of remember claims being made about 2.5 being 10% faster than
2.4/2.3 etc etc. Can anyone say where the speedups were? Presumably we
have a lot of old cruft that could be improved in some way eg moving
loops into comprehensions, using iterator methods etc. Are those sort of
things what we should look at?

Python 2.5 became quite fat. For bare CGI the Python load/init
time eats all improvements. Smaller scripts even loose lot of speed.
I still like Python 2.3 for many other reasons for many
applications - especially for CGI's, on Windows, for deployable
apps, GUI's etc. because the fat coming with Python 2.4 is not
balanced by necessary goods - mostly just fancy things.

What do you mean? Fat of libraries or fat itself?
I tought that 2.5 was faster than precedent versions! :-\

[Robin Becker]

br*****************@gmail.com wrote:

Robin Becker a écrit :
........

AFAIK, most of the speedup comes from optimization of the builtin dict
type, which is the central
data structure in Python. But anyway, as Robert pointed out, using CGI
means lauching
a new Python process for each and every HTTP request.

>Presumably we have a lot of old
cruft that could be improved in some way eg moving loops into comprehensions,
using iterator methods etc. Are those sort of things what we should look at?

Consider profiling your code before doing anything else - unless you're
planning on wasting your time.

our major show stoppers(stuff like stringWidth) are already in C; the remainder
are rather complex methods to do things like paragraph splitting. We're already
considering fcgi to eliminate the startup time. A trial with psyco indicates
another 15% could be obtained there, but I don't think we can use that on the
target platform (which I think is sparc).
--
Robin Becker

[Bruno Desthuilliers]

billie a écrit :

robert wrote

>Robin Becker wrote:

>>A large cgi based web Python-2.3 application needs to be speed improved.
experiments show the following under reasonable testing (these are 2
second reportlab pdf productions)

1) 2.3 --2.5 improvement small 1-2%
2) cgi --fcgi improvement medium 10-12%

I sort of remember claims being made about 2.5 being 10% faster than
2.4/2.3 etc etc. Can anyone say where the speedups were? Presumably we
have a lot of old cruft that could be improved in some way eg moving
loops into comprehensions, using iterator methods etc. Are those sort of
things what we should look at?

Python 2.5 became quite fat. For bare CGI the Python load/init
time eats all improvements. Smaller scripts even loose lot of speed.
I still like Python 2.3 for many other reasons for many
applications - especially for CGI's, on Windows, for deployable
apps, GUI's etc. because the fat coming with Python 2.4 is not
balanced by necessary goods - mostly just fancy things.

What do you mean? Fat of libraries or fat itself?
I tought that 2.5 was faster than precedent versions! :-\

It is. But it takes a bit more time to launch the interpreter. Which is
barely noticiable in most cases, but can be start to be a problem with CGI.

[Gabriel Genellina]

At Wednesday 17/1/2007 10:29, billie wrote:

Python 2.5 became quite fat. For bare CGI the Python load/init
time eats all improvements. Smaller scripts even loose lot of speed.

What do you mean? Fat of libraries or fat itself?
I tought that 2.5 was faster than precedent versions! :-\

It runs faster, but has a slower start time. For CGI, when you launch
a new process with every request, this is bad.
(This is a general problem with CGI itself - perhaps one should
consider other alternatives, like FastCGI, mod_python, WSGI...)
--
Gabriel Genellina
Softlab SRL


__________________________________________________
Preguntá. Respondé. Descubrí.
Todo lo que querías saber, y lo que ni imaginabas,
está en Yahoo! Respuestas (Beta).
¡Probalo ya!
http://www.yahoo.com.ar/respuestas