Bad-mouthing and hostility

[Matt Domsch]

On Wed, Mar 10, 2010 at 03:17:41PM +0100, Thomas Janssen wrote:
> This could change if we provide information about
> downloads/installation of packages (thinking on the one mirror in
> Fedoras hands or information submitted once a month (for example) from
> a users box). I would drop a package if nobody uses it (including me).
> But that's just me.

arguments about statistics/sampling...  We have >240 public mirrors
serving >158Gbits of bandwidth), and >300 private mirrors (bandwidth
unknown).  Most of those have lots more bandwidth than what
Fedora-owned mirrors can provide.  So, if we have a 1 in say 550
chance of hitting a Fedora-owned mirror if all bandwidth were equal;
we weight based on bandwidth of each mirror, so it's actually going to
be larger than that, likely about 1 in 1500.  So at a rough estimate
(modulo user location and available mirrors by country), a package
would need 1500 users for our mirror to serve it up once.  We're
complaining that we can't find 3 users of a single package to provide
feedback.

Sure, we could have a new "quarantine" hunk of the master mirror that
we don't expose to the mirror network as a whole, but only to a
Fedora-owned mirror.  How do we decide which packages to put there?
I'm seeing complexity to gather more information, but if the set of
actions at the end of the day remains "it's up to the maintainer to
keep the package in Fedora or not" - I don't see how the added
information and complexity to gather it would help.

Thanks,
Matt