Branch proxy autoconfig [was Global proxy autoconfig script?]

[Subject updated since the branch configuration issue is rather different
from the global case.]

Oskar Pearson wrote:
> We were talking about a global web server that would run a script, and
> based on the IP address of the client machine, return a different
> java auto-config script.
>...
> > The Proxy Auto Configuration (PAC) script could be generated by a Perl
> We have a related problem:
>
> We have numerous branches spread around the country. We would like to
> get a machine in each of those branches to handle traffic for their
> area, so that connections don't come all the way over our national
> links to get to the cache up here. It would, however, be nice to
> not have 1 different config file for each branch (ie settings would be
> perfectly uniform) I spent a long time sitting and thinking about this,
> and decided that there are basically 4 options:
>
> 1) A central auto-config script that runs a CGI on our machines and
> decides where they are supposed to go based on the request.
>...

At a first glance, this seems like a bad idea, since you don't want all the
client systems fetching their configurations direct from a central web server
over the national links.

However ... if you write the script so it sends an explicit and reasonably
distant expiry timestamp with the script output, making it cacheable, then the
client systems would only go direct to the central web server the first time
they needed a configuration file.

After that, they would typically get it from their local cache except for the
first request after the current copy expired. If the branch cache was down,
then they could still check if they needed an updated config file as long as
the central cache or direct access was defined as a fallback option, so you
would not be stuck with all the clients expecting to use the branch cache and
no way to reconfigure them if it was down for some time (or even taken out of
service permanently...). And the cache could return the most recently cached
version even if the central web server was down when it should have fetched a
new copy, so you would not be totally dependent on the central web server.

Note that getting the script output to be cached would need the script URL to
be chosen so that it did not match any cache configuration rules that would
disable caching (e.g. "cgi-bin" or "?" in the URL, if cache_stoplist is set
as in the Squid sample configuration).

                                John Line

-- 
University of Cambridge WWW manager account (usually John Line)
Send general WWW-related enquiries to webmaster@ucs.cam.ac.uk