Re: introducing http.debian.net
Hello,
On Tue, Jun 26, 2012 at 07:30:58AM -0400, Austin Denyer wrote:
> Donald Norwood wrote:
> > On 06/22/2012 02:24 PM, Darren Baginski wrote:
> >> 22.06.2012, 22:01, "Darren Baginski"<kickbsd@yandex.com>:
> >>>> I've been working on a project that provides an alternative to
> >>>> cdn.d.n. It is based on http redirections.
[..]
> >>> This is very bad idea, instead of having auto mirror selection
> >>> mechanism you've implemented classical Single Point of Failure
> >>> Proper implementation should loop over existing mirrors, starting
> >>> with less latency/faster transfer and then fallback/load balance to
> >>> another,
[..]
> > Can you clarify what you mean by single point of failure? One would
> > think that having a redirector point towards several *matched* mirrors
> > in terms of load balancing, distance, consistency checks, and
> > newer/fresher mirrors would be of great benefit vs. picking a mirror and
> > discovering it is either down, out of sync, off-line, or otherwise
> > befallen. The master list is always available so one can still pick and
> > choose a favored mirror to override the http.d.o suggestion.
>
> I think he means the redirector itself would be a single point of failure.
There would be several redirector instances around the world, and the user
points to the closest redirector using geoDNS.
And as already used in Debian infrastructure, some mini-nagios check ensure the
redirector replies, or remove it from the geoDNS.
--
Simon Paillard
Reply to: