To understand the impact of failed module loads (where Python traverses
sys.path looking for a module), I used the following code:
import string
import random
def random_string(n):
return ''.join(random.choice(string.ascii_letters) for x in range(n))
def random_import():
for i in range(100):
name = random_string(5)+'.'+random_string(5)+'.'+random_string(5)
try:
__import__(name)
except ImportError:
pass
On geom.math:
sage: %timeit random_import()
5 loops, best of 3: 175 ms per loop
On my laptop:
sage: %timeit random_import()
5 loops, best of 3: 47.1 ms per loop
This is both times with warm cache. I also tried enabling the atime mount
option on my laptop, and that did not cause any noticeable changes. I also
tried some background CPU load on my laptop, and as long as I don't
saturate all cores I find only minor (~10%) slowdown. So its mysterious to
me why geom.math is so slow. It would would be interesting to know how
geom.math benchmarks when it is completely idle, right now its at about 50%
CPU load.
On Saturday, August 11, 2012 2:43:36 PM UTC-4, Volker Braun wrote:
>
> I still don't understand why its so slow for you, on a recent machine Sage
> startup with warm cache should be around 1s.
>
> Playing around with the startup a bit, I noticed that the notebook stuff
> imports a lot during startup. Why do we need to import anything from the
> notebook except the notebook() function? This seems wasteful. Cutting out
> the notebook imports saves me about 1/4 of the startup time (from ~1s to
> 0.75s).
>
> We also have a lot of stuff in the sys.path that causes many failed open()
> calls upon startup as Python has to check every directory for modules.
> About half of all are notebook-related python packages that have no use in
> mathematics. These are also subdirectories of site-packages so they are
> searched twice since site-packages is also is sys.path. Just getting rid of
> packages that are only used in the notebook saves about a third of all
> open() calls during startup. But on my laptop that only saves something on
> the order of 0.01s, basically nothing. But if you have a slow file system
> then it might make a difference.
>
>
>
>
> On Friday, August 10, 2012 8:51:17 PM UTC-4, William wrote:
>>
>> Hi,
>>
>> I'm sad that on a fast SSD repeatedly doing the following:
>>
>> wstein@geom:/usr/local/sage/sage-5.2$ time echo "" | ./sage
>> ----------------------------------------------------------------------
>> | Sage Version 5.2, Release Date: 2012-07-25 |
>> | Type "notebook()" for the browser-based notebook interface. |
>> | Type "help()" for help. |
>> ----------------------------------------------------------------------
>> sage: sage:
>> Exiting Sage (CPU time 0m0.03s, Wall time 0m0.03s).
>>
>> real 0m1.902s
>> user 0m1.428s
>> sys 0m0.424s
>>
>> always outputs 1.9 seconds! That test was < 1 second on sage.math
>> when I put in a doctest to make sure it didn't exceed that 1 second,
>> when we were trying to improve the sage startup time. In the
>> intervening year (?), the test is ignored or gone, and the startup
>> time seems to have got a bit worse. Dang.
>>
>> On another faster machine, I'm getting:
>>
>> wstein@combinat:/usr/local/sage/sage-5.2$ time echo "" | ./sage
>> ----------------------------------------------------------------------
>> | Sage Version 5.2, Release Date: 2012-07-25 |
>> | Type "notebook()" for the browser-based notebook interface. |
>> | Type "help()" for help. |
>> ----------------------------------------------------------------------
>> sage: sage:
>> Exiting Sage (CPU time 0m0.04s, Wall time 0m0.05s).
>>
>> real 0m1.580s
>> user 0m1.224s
>> sys 0m0.356s
>>
>>
>> Any ideas?
>>
>>
>>
>>
>> William
>>
>>
>>
>>
>> --
>> William Stein
>> Professor of Mathematics
>> University of Washington
>> http://wstein.org
>>
>
--
--
To post to this group, send an email to sage-devel@googlegroups.com
To unsubscribe from this group, send an email to
sage-devel+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org
