[Python-Dev] bigmemtests for really big memory too slow

Antoine Pitrou solipsis at pitrou.net
Tue Sep 6 16:46:37 CEST 2011


For the record, I've disabled automatic builds on the bigmem buildbot
until things get sorted out a bit (no need to eat huge amounts of RAM
and eight hours of CPU each time a commit is pushed, only to have the
process killed :-)). It's still possible to run custom builds, of
course.

Regards

Antoine.



On Tue, 06 Sep 2011 15:03:32 +0200
martin at v.loewis.de wrote:
> I benchmarked some of the bigmemtests when run with -M 80G. They run really
> slow, because they try to use all available memory, and then take a lot of
> time processing it. Here are some runtimes:
> 
> test_capitalize (test.test_bigmem.StrTest) ... ok (420.490846s)
> test_center (test.test_bigmem.StrTest) ... ok (149.431523s)
> test_compare (test.test_bigmem.StrTest) ... ok (200.181986s)
> test_concat (test.test_bigmem.StrTest) ... ok (154.282903s)
> test_contains (test.test_bigmem.StrTest) ... ok (173.960073s)
> test_count (test.test_bigmem.StrTest) ... ok (186.799731s)
> test_encode (test.test_bigmem.StrTest) ... ok (53.752823s)
> test_encode_ascii (test.test_bigmem.StrTest) ... ok (8.421414s)
> test_encode_raw_unicode_escape (test.test_bigmem.StrTest) ... ok (3.752774s)
> test_encode_utf32 (test.test_bigmem.StrTest) ... ok (9.732829s)
> test_encode_utf7 (test.test_bigmem.StrTest) ... ok (4.998805s)
> test_endswith (test.test_bigmem.StrTest) ... ok (208.022452s)
> test_expandtabs (test.test_bigmem.StrTest) ... ok (614.490436s)
> test_find (test.test_bigmem.StrTest) ... ok (230.722848s)
> test_format (test.test_bigmem.StrTest) ... ok (407.471929s)
> test_hash (test.test_bigmem.StrTest) ... ok (325.906271s)
> 
> In the test suite, we have the bigmemtest and precisionbigmemtest
> decorators. I think bigmemtest cases should all be changed to
> precisionbigmemtest, giving sizes of just above 2**31. With that
> change, the runtime for test_capitalize would go down to 42s.
> 
> What do you think?
> 
> Regards,
> Martin
> 
> 
> 




More information about the Python-Dev mailing list