Re: periodic daily takes a very long time to run (14-stable)

On 29/10/23 00:58, void wrote:
> On Sun, Oct 29, 2023 at 12:49:14AM +0200, Guido Falsi wrote:
> 
>> Correcting myaelf:
>>
>> The version we currently have in the ports tree (3.7.x) does not care 
>> for atime, as far as I can see in the source code. It checks mtime only.
>>
>> Newer versions do use it though!
> 
> Thanks for checking. I've set it to on and think I'll leave it on as
> long as there's no e.g performance penalty. One less thing to think about
> changing again when the port moves versions.

Well actually atime does have a performance penalty. The fact has been 
discussed a lot, but the net effect of atime is that every pure read 
operation causes a write operation (to update atime) to be created, so 
it does cause extra disc activity.

There was an old thread involving Linux Torvalds about this. The thread 
is about the linux kernel, but most arguments apply to any filesystem 
saving last access time [1]

That's why since a few years ago it is being disabled by default almost 
everywhere. Enabling it for all filesystems makes no sense. Enabling it 
for ccache, depends, my wild guess is that the performance penalty is 
really minor, but I have performed no testing.

[1] 
http://web.archive.org/web/20110427023154/http://kerneltrap.org/node/14148

-- 
Guido Falsi <mad@madpilot.net>