On Wed, 29 Apr 2020 at 20:22, Chuck Anderson <cra(a)wpi.edu> wrote:
On Wed, Apr 29, 2020 at 10:50:00PM +0200, Adrian Reber wrote:
> On Wed, Apr 29, 2020 at 04:44:08PM -0400, James Cassell wrote:
> >
> > On Wed, Apr 29, 2020, at 1:31 PM, Theo Morra wrote:
> > > Heya,
> > >
> > > We're adding some more storage to our mirrors - how big is buffet?
> > >
> >
> > Someone said 3.8 TB the other day.
>
> Unfortunately I was off by one. It is 38TB. Knowing that, my proposal to
> require mirroring fedora-buffet for new tier 1 mirrors sounds not so
> good anymore as that is really a lot.
It is a lot, but I think it should still be required. Tier1 should
have everything any Tier2 might want to pull a subset of.
In the past I would agree, but we (aka Fedora) had to drop mirroring
archives from 2 of our systems (download-ib01/download-i2 and
download-cc-rdu01) because we didn't have the disk space on these
servers to do that much storage. I am not sure trying to shove every
use case for a top mirror into one box is going to work anymore.
At the moment I only see the following mirrors offering archives via
mirrormanager:
mirror.math.princeton.edu
pubmirror1.math.uh.edu
d2lzkl7pfhq30w.cloudfront.net # a fedora project caching server
dl.fedoraproject.org
I expect there are probably private ones which I am not seeing via my
simple curl |& grep but the majority of users who are grabbing stuff
from archived data are only getting it from those 4 places. While it
would be good for more, I can also see that the majority of mirror
traffic is EPEL driven with only the latest 3 Fedora's taking up the
majority of users. Having those fed fast is probably more important
than Fedora 18.
Would a different tiering system work better in the same way as we
have different rsync targets?
--
Stephen J Smoogen.