On Wed, Jun 21, 2017 at 02:04:50PM -0400, Matthew Miller wrote:
On Wed, Jun 21, 2017 at 10:46:26AM -0500, Dennis Gilmore wrote:
> There is a fairly large amount of people who continue to use a release
> after it is EOL. It was about making sure that the archives do not get
> overloaded with requests as there is not many mirrors there. and to
> allow for a smooth transition phase for users. Perhaps we could change
> it and keep one less release. I would like to see how many hits we get
> on mirrormanager for f22 and f23 today so we can make it an informed
> decision.
So, the good news on this point is that we're doing a relatively good
job of keeping people current -- the work we're doing on making
upgrades smooth seems to be paying off. A little shy of 47% of mirror
checkins come from F25 (with F26 beta and Rawhide contributing another
1.5%). F24 is 15%, and F23 is 12%.
F22 drops down to 6.2%, F21 is 4.3%, and F20 is 5.9%. (F20 looks like
it's going to linger for a good long time.)
Everthing below F20 is basically peanuts in comparison (although
shout-out to F8, still holding on at around 1.6%, which puts it higher
than every pre-F20 release).
I'm not sure exactly where the cutoff should be, but I guess I'd be a
bit hesitant to move F23 away just yet. But, maybe we should make it
easier for mirrors to have only-current and current-1? (F24 is only
15%, but of course that was much higher six months at F25 launch.)
I think it would be OK to have 3 releases always on mirrors.
Removal of current-2 would be nice when new release is comming
(week or month before release).
When F22 can be removed from mirrors? It's 300 GB of release,
170 GB of updates and 13 GB of testing updates.
In summary aprox. 0.5 TB.
SAL