Talloc speed and pool usage
Pepijn de Vos
pepijndevos at gmail.com
Sat Aug 4 11:08:23 MDT 2012
Hi,
I am writing this persistent hashmap, and up to a few days it used its
own reference counting with malloc. I'm now trying to use talloc instead.
In terms of code size, it was an improvement: *96 additions* and *159
deletions*.
In terms of speed, it was a major regression. It said in he docs that it
would have a few percent slowdown. My test went from 15 seconds to 5
minutes. The only thing I can think of is that I'm relying heavily on
reference counting.
Is it just that slow, or am I using it wrong? My change:
https://github.com/pepijndevos/perper/commit/c9a05b91036bc5df1218db2b1e77dc9d180f37be
I found that you have pool allocations to speed things up. I tried to
use them in my test, but it aborts when I use them.
https://github.com/pepijndevos/perper/blob/talloc/test.c
You can see all this for yourself by running "make && time ./test" using
either the master or talloc branch and possibly uncomment my pool usage.
Regards,
Pepijn de Vos
http://pepijndevos.nl
More information about the samba-technical
mailing list