Re: Potential regression: high server-side memory consumption during import

Stefan Sperling <stsp_at_elego.de> writes:

> On Fri, Mar 02, 2018 at 06:20:48PM +0000, Philip Martin wrote:
>>
>> Yes, that solves the memory use problem.
>
> Nice, I'll commit it then. This might not be a final fix but at least
> it's a step forward.
>
>> There is still a time penalty:
>>
>> Your patch:
>>
>> $ time svn log -q http://localhost:8888/repo/asf/subversion > /dev/null
>>
>> real 0m7.695s
>> user 0m0.429s
>> sys 0m0.283s
>>
>> apache httpd cpu: 7.30
>>
>> Revert r1779188,1778923
>>
>> $ time svn log -q http://localhost:8888/repo/asf/subversion > /dev/null
>>
>> real 0m5.102s
>> user 0m0.460s
>> sys 0m0.250s
>>
>> apache httpd cpu: 5.10
>
> If I understand it correctly, that's because 1778923 disabled caching
> on a per-connection basis? I don't really understand the log message.

For comparison, 1.9 gives

 $ time svn log -q http://localhost:8888/repo/asf/subversion > /dev/null

 real 0m10.044s
 user 0m0.404s
 sys 0m0.064s

 apache httpd cpu: 9.98

1.10 would be nearly twice as fast as 1.9 but rereading the authz
removes over half of the gain.

All of those figures are the first run after starting Apache, i.e. when
the OS cache and the Subversion cache is cold. With a hot Subversion
cache the values are:

            1.9: 6.0s
   patched 1.10: 5.5s
  reverted 1.10: 3.3s

Again, 1.10 would be nearly twice as fast, but now rereading the authz
removes most of that gain.

-- 
Philip