[packages/rpm] - disable removing duplicate ldconfig invocations optimization http://rpm5.org/community/rpm-devel

Jeffrey Johnson n3npq at me.com
Fri Sep 28 03:32:44 CEST 2012


On Sep 13, 2012, at 2:22 AM, Elan Ruusamäe <glen at pld-linux.org> wrote:

> On 09/10/2012 04:07 PM, Jeffrey Johnson wrote:
>> On Sep 10, 2012, at 3:52 AM, Elan Ruusamäe wrote:
>> 
>>> On 10.09.2012 09:12, baggins wrote:
>>>> commit b6aa06669d52acb5791a88eef2b8499391c5b44e
>>>> Author: Jan Rękorajski<baggins at pld-linux.org>
>>>> Date:   Sun Sep 9 19:18:20 2012 +0200
>>>> 
>>>>     - disable removing duplicate ldconfig invocations optimization
>>>>       http://rpm5.org/community/rpm-devel/5380.html
>>> just to comment why optimization is not needed:
>>> recent glibc caches ldconfig calls, so it's not that heavy, and besides fs cache should be hot as well
>>> 
>> Subjective claims of "not that heavy" and "should be hot" do not support the
>> conclusion "optimization is not needed" without measurements.
>> 
>> Can you provide measurements?
> cold, without cache:
> # rm -f /var/cache/ldconfig/aux-cache
> # time ldconfig
> 
> real    0m15.335s
> user    0m0.063s
> sys     0m0.160s
> 
> hot, without cache:
> # rm -f /var/cache/ldconfig/aux-cache
> # time ldconfig
> 
> real    0m0.068s
> user    0m0.030s
> sys     0m0.037s
> 
> # rm -f /var/cache/ldconfig/aux-cache
> # time ldconfig
> 
> real    0m0.067s
> user    0m0.040s
> sys     0m0.027s
> # rm -f /var/cache/ldconfig/aux-cache
> # time ldconfig
> 
> real    0m0.068s
> user    0m0.037s
> sys     0m0.027s
> 

Thank you for measurements.

What I see is I/O to initially read libraries is expensive,
regenerating the index is cheap. Re-reading loaded libraries
costs almost nothing.

So removing duplicate ldconfig calls in RPM matters hardly at all.

Agree?

73 de Jeff



More information about the pld-devel-en mailing list