On Sat, Sep 20, 2008 at 04:49:24PM +0100, Peter Tribble wrote:

> On Tue, Sep 16, 2008 at 6:03 PM,  <[EMAIL PROTECTED]> wrote:
>
> > The goal is to have the download front end be interchangable.  You
> > could use Apache, CherryPY, your HTTPd of choice, or even FTP or file.
> 
> As I understand it, that's only for the data part. You still have to have
> a primary server, which doesn't solve the main problem.

No, the goal is to have all the non-publication client operations except
for search be completely HTTP/1.1 and a bunch of GETs (or some other
protocol).  The first phase out of that will be to transition the filelist
operation into a bunch of pipelined file operations, which we're hoping to
be able to do with pycurl.

That, combined with the mirroring which is already in place, will allow any
dumb webserver (well, one with a simple rewrite mechanism, but we can
probably even get rid of that, if necessary) to serve all the
bandwidth-hogging content.

It will also allow for manifests and catalogs to be downloaded simply.
They already are, actually, it's just that mirroring treats them
differently, so they seem special in that respect.  Once you've done that,
you've no need of the authority's origin.

Danek
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to