Bug#212732: support redirects and interactive authentication (Progeny)
On Sat, Mar 20, 2004 at 12:51:01AM +0100, Wichert Akkerman wrote:
> Previously Rico -mc- Gloeckner wrote:
> > i wouldnt recommend anycast with tcp-sessions.
>
> Why? For long-lasting sessions I can see why, but for normal http
> traffic which is just a bunch of short-lived sessions anycast should
> work great.
Your 56k User who just tries to download that one large package will thank you.
('get -r' (ie reget) may be of help here, though)
I have still not seen any data on how interrupting anycast to TCP is, so
i will refrain from using Anycast for mirroring files. Additionally we
face two problems:
- you have no idea wether your current reachable server is a real
mirror or wether someone is currently injecting a mirror which looks
like a debian mirror but contains cracked debs.
- Not every mirror has the possibility to influence the bgp table. At
all, i'd still like to give those mirrors the possibility to do their
task in mirroring.
Yet, iam not saying that my method is the way to go -- i had various
ideas to patch apt based on dns, but for the optimum of redirection, dns
would probably need not to be cached, though i would like to collect
numbers on how good the approximation is, so id take that opportunity
too, if someone dares to patch apt for that. (Iam not a C++ coder at
all, really).
I am also open to any suggestions and discussion which may result in a
good-fashioned way to distribute users over mirrors with the aim to make
it as easy as possible for them.
--
Rico -mc- Gloeckner | 1024D/61F05B8C | jabber:mc@micq.org
http://www.ukeer.de | RICO-RIPE | sip:42@daheim.ukeer.de
== mv ~/.signature http://www.ukeer.de/signature.html ==
Reply to: