On Thu, Feb 6, 2020 at 12:01 AM Kevin Kofler <kevin.kofler@chello.at> wrote:
Oh, and to answer your other point:

Lukas Brabec wrote:
> It is pretty common for us in Fedora QA (well, I'm quite biased in this
> case).
> And no, we cannot compose our images, we have to test the exact same
> images that will be shipped. We cannot test custom images and pretend
> the results are going to be the same for official images.

So you propose to optimize Fedora for your own internal use at the expense
of thousands of users?

        Kevin Kofler

Assuming that your numbers are even accurate (which I have not seen any proof of so far), your vaguely defined "thousands" (semantically implies < 10, but absolutely < 20k, especially since you tend to use hyperbole, and that would definitely be "tens of thousands" to suit your case better) are _at worst_ less than 2%, but most probably less than 1% of _downloads_ [1].
Given that at least some users perform more than one installation per downloaded image, it is quite probable that it's even smaller percentage of _users_, But let's be generous and say it's 5% of users, that really struggle, and currently download a Fedora image for lets call it "tens of hours" (when "a few percent increase or decrease in download  time can mean hours of difference", it is quite obvious that the base download time is enormous to start with). If there are optimizations to be performed on the project's scale, what would you say has a bigger impact:
 a) optimization targeting < 5%
 b) optimization targeting > 95%

Looking forward to seeing your proof based on hard-data,

Josef

[1] https://www.pcworld.com/article/3038353/fedora-project-leader-matthew-miller-reveals-whats-in-store-for-fedora-in-2016.html