www-infrastructure-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Peter Poeml <po...@cmdline.net>
Subject Re: mirrorbrain
Date Fri, 03 Jul 2009 15:57:23 GMT
Hi!

To quickly introduce myself, I'm the author of MirrorBrain and I noticed
your discussion here.

On Tue, May 12, 2009 at 07:21:37 -0400, Philip M. Gollucci wrote:
> Paul Querna wrote:
> >thought some people might find this interesting:
> >http://mirrorbrain.org/
> >
> >as an alternative to mirmon.
> >
> >Not that I want to switch, mirmon works great[1], the only thing
> >mirrorbrain seems to do better is the ability to crawl the mirrors and
> >selectively check the checksums of files, making sure the mirrors are
> >accurate -- right now all we do is a date check with mirmon.
> >
> >[1] - http://www.apache.org/mirrors/
> 
> How much effort do you think is involved in switching.
> 
> ASF Mirroring is something I've been meaning to look into.

Mirmon is a little different than MirrorBrain. It mainly checks
timestamps on the mirrors, keeps a history and visualizes it. As far as
I understand, that's all that it does. I know of the closer.cgi that you
guys use to find mirrors, and I don't know whether it's part of mirmon
or not. (I suppose that it runs based on the mirror list of mirmon.)

You could replace closer.cgi with MirrorBrain. You could (and would)
keep mirmon for its nice visualiziation, and generate a mirmon config
out of MirrorBrain's mirror database. You'd maintain the mirror metadata
within MirrorBrain's database because it's more powerful; generating a
mirmon config would be a trivial task.

I know the page http://www.apache.org/dyn/closer.cgi and it would easily
be possible to generate the same page from MirrorBrain, so you could
keep that page for convenience of users to be able to pick mirrors
manually.

At the same time, MB would redirect transparently to all files known to
be on mirrors through a single URL space on your download server, which
means that all your projects would have the option to publish direct
links to their releases. They could still also provide the link to the
closer.cgi page, for a manual choice, for the odd cases where it's
needed.

MB can be configured to deliver signature (*.asc) files and the like
directly, so users are never sent to mirrors for those files.

Looking at the closer.cgi page, the list of mirrors is rather short,
at least a little shorter than I would expect - in the past, I got the feeling
that there are many more mirrors that mirror subtrees of the tree, for
example the httpd tree - could that be the case? Mirrors don't
need to be "complete" mirrors for MirrorBrain, so you would maybe be
able to make use of more mirrors.

I would be very happy to help you getting a MB instance up and running
on one of your download servers, if you are interested! (This could also
be done on a mirror, to try it out, any machine which has the file tree.)

I think it would be very valuable for both of us to join forces on this.
Handling a mirror infrastructure is intricate business and there is so
much too learn. There's more to share than software - the expertise in
how content delivery in different parts of the world works best is not
easily acquired. In my views, it would make a lot of sense if we
content creators and the content deliverers (mirror admins) cooperate
more. The goal of the MirrorBrain project is to be a basis for this.

If you want to know more, please let me know. I'm subscribed to the
infrastructure-dev mailing list, and will stay for a little while, but
it would be safe to reach me at my email, or you could just post to the
mirrorbrain at mirrorbrain.org mailing list. (You don't need to be
subscribed to post.)

Thanks,
Peter

Mime
View raw message