Re: [Python-Dev] Benchmarks: Comparison between Python 2.7 and Python 3.6 performance

2016-11-07 Thread Wolfgang Maier

On 05.11.2016 10:56, Antoine Pitrou wrote:


Hi Victor,

On Fri, 4 Nov 2016 13:53:10 +0100
Victor Stinner  wrote:


Raw results of Python 3.6 compared to Python 2.7:


That's interesting, but I would be personally more interested in
a performance comparison of 3.5 and 3.6, to know if anything
interesting (or worrying :-)) has happened there.



You can get this as well from https://speed.python.org/comparison/
and https://speed.python.org/timeline and looking at this, I think there 
is something worrying indeed:
Startup time has increased by ~ 30 % between 3.5 and 3.6 again. More 
specifically, all this increase happened between Sep 09 and Sep 15.


I have no clue why that is, but it is definitely the biggest effect far 
and wide.



The performance differences between 2.7 and 3.x are quite well-known by
now, and none of them are really dramatic except for the increase in
startup time.

Regards

Antoine.




___
Python-Dev mailing list
[email protected]
https://mail.python.org/mailman/listinfo/python-dev
Unsubscribe: 
https://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com


Re: [Python-Dev] Benchmarks: Comparison between Python 2.7 and Python 3.6 performance

2016-11-07 Thread Guido van Rossum
Re:
https://speed.python.org/timeline/#/?exe=4&ben=python_startup&env=1&revs=50&equid=off&quarts=on&extr=on

That's suspiciously close to the core sprint. Since the -S time stayed
roughly the same I suspect that either a new module was added to the
startup sequence or one of the (too many) modules already involved grew a
lot. My money is on a new module. Using `python -v -c pass` on a Python
built from the Sept. 9 tree and one built from Sept. 15, it shouldn't be
too hard to figure out which new module(s).

On Mon, Nov 7, 2016 at 7:11 AM, Wolfgang Maier <
[email protected]> wrote:

> On 05.11.2016 10:56, Antoine Pitrou wrote:
>
>>
>> Hi Victor,
>>
>> On Fri, 4 Nov 2016 13:53:10 +0100
>> Victor Stinner  wrote:
>>
>>>
>>> Raw results of Python 3.6 compared to Python 2.7:
>>>
>>
>> That's interesting, but I would be personally more interested in
>> a performance comparison of 3.5 and 3.6, to know if anything
>> interesting (or worrying :-)) has happened there.
>>
>>
> You can get this as well from https://speed.python.org/comparison/
> and https://speed.python.org/timeline and looking at this, I think there
> is something worrying indeed:
> Startup time has increased by ~ 30 % between 3.5 and 3.6 again. More
> specifically, all this increase happened between Sep 09 and Sep 15.
>
> I have no clue why that is, but it is definitely the biggest effect far
> and wide.
>
>
> The performance differences between 2.7 and 3.x are quite well-known by
>> now, and none of them are really dramatic except for the increase in
>> startup time.
>>
>> Regards
>>
>> Antoine.
>>
>>
>>
> ___
> Python-Dev mailing list
> [email protected]
> https://mail.python.org/mailman/listinfo/python-dev
> Unsubscribe: https://mail.python.org/mailman/options/python-dev/guido%
> 40python.org
>



-- 
--Guido van Rossum (python.org/~guido)
___
Python-Dev mailing list
[email protected]
https://mail.python.org/mailman/listinfo/python-dev
Unsubscribe: 
https://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com


[Python-Dev] Installing Python on Windows

2016-11-07 Thread Steve Dower

Hi all

Those of you who follow me on Twitter (@zooba) may have noticed that I 
posted one of my rare blog posts over the weekend about the increasing 
range of Python installers available for Windows.


I figured I'd draw some attention here in case others are interested in 
my rationale for why the main installer is how it is, and why I'm also 
experimenting with other forms of installer or package:


http://stevedower.id.au/blog/why-so-many-python-installers/

Arguably this would make a good section in the docs or a PEP, though 
it's very point-in-time content, so a blog seems fine for now. There's 
also not really anything in the docs about how Linux distributors 
package up CPython, so maybe it's best as entirely standalone content.


Cheers,
Steve

(In case anyone is suspicious, I earn no income from my blog and there 
are no ads. So I hope I'm not promoting the link for selfish reasons :) )

___
Python-Dev mailing list
[email protected]
https://mail.python.org/mailman/listinfo/python-dev
Unsubscribe: 
https://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com


Re: [Python-Dev] Benchmarks: Comparison between Python 2.7 and Python 3.6 performance

2016-11-07 Thread Victor Stinner
2016-11-07 20:20 GMT+01:00 Guido van Rossum :
> Re:
> https://speed.python.org/timeline/#/?exe=4&ben=python_startup&env=1&revs=50&equid=off&quarts=on&extr=on
>
> That's suspiciously close to the core sprint. Since the -S time stayed
> roughly the same I suspect that either a new module was added to the startup
> sequence or one of the (too many) modules already involved grew a lot. My
> money is on a new module. Using `python -v -c pass` on a Python built from
> the Sept. 9 tree and one built from Sept. 15, it shouldn't be too hard to
> figure out which new module(s).

I identified the regression, I created:

   http://bugs.python.org/issue28637

It's a regression caused by:

   https://hg.python.org/cpython/rev/223731925d06/
   http://bugs.python.org/issue28082

The change added "import enum" in Lib/re.py. I propose to revert this
change for now, and discuss later a solution which doesn't impact
performances.

It's nice to see that my work of performance was useful to catch a
performance regression.

Victor
___
Python-Dev mailing list
[email protected]
https://mail.python.org/mailman/listinfo/python-dev
Unsubscribe: 
https://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com


[Python-Dev] Benchmarks: Comparison between Python 3.5 and Python 3.6 performance

2016-11-07 Thread Victor Stinner
Hi,

Antoine Pitrou asked me to compare Python 3.6 to Python 3.5. Here are
results on the speed-python server using LTO compilation (but not
PGO):

$ python3 -m perf compare_to 2016-11-03_15-37-3.5-89f7386104e2.json
2016-11-03_15-38-3.6-c4319c0d0131.json -G --min-speed=5
Slower (17):
- call_method_slots: 16.7 ms +- 0.2 ms -> 28.3 ms +- 0.7 ms: 1.70x slower
- call_method: 16.9 ms +- 0.2 ms -> 28.6 ms +- 0.8 ms: 1.69x slower
- call_method_unknown: 18.5 ms +- 0.2 ms -> 30.8 ms +- 0.8 ms: 1.67x slower
- python_startup: 20.3 ms +- 0.7 ms -> 26.9 ms +- 0.6 ms: 1.33x slower
- regex_compile: 397 ms +- 4 ms -> 482 ms +- 6 ms: 1.21x slower
- mako: 42.2 ms +- 0.5 ms -> 49.7 ms +- 2.5 ms: 1.18x slower
- deltablue: 17.8 ms +- 0.2 ms -> 20.1 ms +- 0.6 ms: 1.13x slower
- chameleon: 29.1 ms +- 0.4 ms -> 31.9 ms +- 0.4 ms: 1.10x slower
- genshi_text: 89.8 ms +- 4.2 ms -> 97.8 ms +- 1.1 ms: 1.09x slower
- pickle_pure_python: 1.30 ms +- 0.02 ms -> 1.41 ms +- 0.03 ms: 1.08x slower
- logging_simple: 35.8 us +- 0.7 us -> 38.4 us +- 0.7 us: 1.07x slower
- sqlalchemy_declarative: 331 ms +- 3 ms -> 354 ms +- 6 ms: 1.07x slower
- logging_format: 42.6 us +- 0.5 us -> 45.5 us +- 0.8 us: 1.07x slower
- call_simple: 13.1 ms +- 0.4 ms -> 13.9 ms +- 0.2 ms: 1.06x slower
- genshi_xml: 197 ms +- 2 ms -> 209 ms +- 2 ms: 1.06x slower
- richards: 190 ms +- 5 ms -> 201 ms +- 6 ms: 1.06x slower
- go: 607 ms +- 19 ms -> 640 ms +- 26 ms: 1.05x slower

Faster (15):
- xml_etree_iterparse: 498 ms +- 11 ms -> 230 ms +- 5 ms: 2.17x faster
- unpickle_list: 10.6 us +- 0.3 us -> 7.86 us +- 0.16 us: 1.35x faster
- xml_etree_parse: 345 ms +- 6 ms -> 298 ms +- 8 ms: 1.16x faster
- xml_etree_generate: 362 ms +- 5 ms -> 320 ms +- 8 ms: 1.13x faster
- scimark_lu: 589 ms +- 24 ms -> 523 ms +- 18 ms: 1.13x faster
- regex_effbot: 5.88 ms +- 0.06 ms -> 5.23 ms +- 0.05 ms: 1.13x faster
- sympy_sum: 269 ms +- 7 ms -> 244 ms +- 7 ms: 1.10x faster
- spectral_norm: 317 ms +- 6 ms -> 287 ms +- 2 ms: 1.10x faster
- sympy_expand: 1.25 sec +- 0.04 sec -> 1.15 sec +- 0.03 sec: 1.09x faster
- xml_etree_process: 291 ms +- 5 ms -> 268 ms +- 14 ms: 1.09x faster
- float: 336 ms +- 8 ms -> 310 ms +- 7 ms: 1.08x faster
- regex_dna: 323 ms +- 3 ms -> 298 ms +- 3 ms: 1.08x faster
- unpickle: 38.4 us +- 0.6 us -> 35.5 us +- 1.2 us: 1.08x faster
- crypto_pyaes: 267 ms +- 3 ms -> 249 ms +- 2 ms: 1.07x faster
- nbody: 262 ms +- 3 ms -> 246 ms +- 3 ms: 1.07x faster

Benchmark hidden because not significant (31): 2to3, chaos, (...)
---

call_method_* slowdown: http://bugs.python.org/issue28618

python_startup slowdown: http://bugs.python.org/issue28637

I didn't analyze results yet.

Victor
___
Python-Dev mailing list
[email protected]
https://mail.python.org/mailman/listinfo/python-dev
Unsubscribe: 
https://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com