- From: Joel N. Weber II <nemo@koa.iolani.honolulu.hi.us>
- Date: Mon, 9 Dec 1996 13:44:50 -1000 (HST)
- To: Koen Holtman <koen@win.tue.nl>
- cc: Brian Morin <bmorin@WPI.EDU>, www-talk@w3.org
On Tue, 10 Dec 1996, Koen Holtman wrote:
> >Is there anything I'm overlooking?
>
> It is not easy for a client to quickly determine which mirror which offers
> the highest bandwidth. A HTTP protocol element for sending a list of
> mirrors to the client only solves the easy part of the problem.
>
> I believe that there can already be some mirror negotiation when resolving a
> hostname to an IP address, but I don't know how effective that is.
Yes, I think that's the key problem.
The Linux mailing lists are distributed on different servers depending
on your domain name. That helps, but it isn't perfect.
I have gotten mail that went from vger.rutgers.edu to nic.funet.fi
to koa.iolani.honolulu.hi.us. Someone must think that's an efficient
route if that's how it's configured, but I don't see the logic there.
OTOH, most users don't know what's optimal either, and time the user
spends picking a server is a ridiculous waste of a user's time.
Maybe proxies do a better job solving this if you design the proxy network
correctly. In other words, for the most efficient possible performance,
there would be a proxy server in the iolani.honolulu.hi.us domain, another
one at mhpcc.com, etc. Right now there's just a proxy for Iolani which
sends the request directly to the server.
nemo
http://www.cyclic.com/~nemo
<nemo@koa.iolani.honolulu.hi.us> <devnull@gnu.ai.mit.edu>
----------------------------------------------------------------------------
"...For I have not come to call the righteous, but sinners." -- Mathew 9:13
Received on Monday, 9 December 1996 18:47:21 UTC