On 01/02/18 14:03, Stuart Henderson wrote:
> Hosting a large git repository is not trivial, it uses far more server
> resources (memory and cpu time) than an anoncvs/cvsync/rsync mirror, and
> OpenBSD src/ (or even just ports/) is *huge* for a git repo. It works
> better on Linux where things are more separated. Even *just the kernel*
> is split across multiple repos.
The Linux kernel repo is multiple times the size of OpenBSD-src [1],
so I don't see how things being more separated helps them re: hosting.
Perhaps kernel.org just has more hardware to throw at the problem?
And in case anyone else was confused, the Linux kernel itself isn't split
across multiple repos: you can build a fully functional one from a single
checkout. It is the kernel *development* that is split across multiple
repos, with occasional merges to mainline.
[1]: Naive estimate based on comparing object counts when cloning from
GitHub:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/ -
5,779,337 objects,
https://github.com/openbsd/src - 1,741,047 objects.
> Anyway, has anyone fetched your openbsd-src0-test repo from github while
> crossing crypto export boundaries? That has the exact same issue,
> except that now as it's your repo, it may well be considered that it's
> *you* that is responsible for exporting it.
Surely the responsibility for exporting lies with the one doing the
checkout? Otherwise I don't see how operators of OpenBSD CVS mirrors
in the US aren't in the same position.
Or is there some technical distinction between "mirroring" and "checking out"
a repository? (I ask because the warning against fetching sources from USA
when located outside North America only appears on
https://www.openbsd.org/cvsync.html, not https://www.openbsd.org/anoncvs.html
or https://www.openbsd.org/ftp.html.)
