User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.5.0
If you're only getting a handful of
downloads each day, I think it's safe to put those files on the
archives.
Our bandwidth usage will always go up
regardless -- and since mirrors handle about 95% of our download
requests, we don't want to impose too much on them.
Thanks for everyone's quick action.
Denis
On 2020-04-15 8:21 p.m., Jonah Graham
wrote:
Hello Denis,
(please feel free to answer publicly if you
think this is generally useful)
Can I get some guidance on whether
something is worth leaving on the download and therefore
mirrored or not? I assume there is a tradeoff of mirror
storage vs EF bandwidth usage, or similar.
For example, for EPP each download file is
~100-400MB, for Neon there are 70 such files, 34 of
which have < 1 download per day. With the most
downloaded file (eclipse-jee-neon-3-win32-x86_64.zip)
being almost 700 per day*.
So how many of these downloads per day
make a file worth having mirrored? If the answer is <
1 download per day is not worth mirroring, then I could
potentially transfer between 40 and 80 GB to archive for
EPP.**
The above questions all assume that
"Older releases must be moved to archive.eclipse.org."
has a non-specific definition for "Older".
** How much I save depends a lot on how
much time I spend identifying which specific files to
archive as some files (JEE specifically) in each release
still gets downloaded a lot.
Moving P2 repos to archive is unhelpful for a user
who attempts a Maven build from the archive repo.
The first Bug above is a Tycho fix for the original
OCL Bug report. Alternatively the EF could ignore a
non-archive p2.mirrorsURL so that the archived repo
behaves just as if the archiving had stripped the
p2.mirrorsURL.
Regards
Ed Willink
On 13/04/2020 14:51, Denis Roy wrote:
Greetings,
As download.eclipse.org
is mirrored worldwide, is it important to keep
your download footprint to a minimum.
Most of our mirror sites are also host to a
number of OSS projects, and enterprise disk
space is not cheap. Some strategies to note:
- Nightly, weekly,
milestone and RC builds should be deleted
- Older releases must be
moved to archive.eclipse.org.
If moved into the same directory structure,
links typically do not need to be updated, as
our 404 Not Found handlers will look for the
file in the archives before throwing the error.
The risk of not cleaning
up is significant. Mirrors, looking to liberate
disk space, will simply purge the worst
offenders, and with fewer mirror sites offering
Eclipse bits, our servers must handle more
downloads, which means unnecessary expense for
the Eclipse Foundation.