Unidentified subject!

Henrik Nordstrom <hno@hem.passagen.se> wrote:
>Simon Rainey wrote:
>> What I'm concerned about is this. Suppose our transatlantic links go
>> down (we're based in the UK) so we lose all US connectivity. Now maybe
>> 80% of requests are for US sites. If Squid (or the dnsserver) is going
>> to block because it can't reach those sites then it will eventually
>> grind to a halt
>
> Now, that is a problem. Squid is an application gateway, and as such it
> is very sensitive to network outage. The main problem is that the number
> of pending TCP connections quickly build up when a major host is
> unreachable. There are a couple of things you may do to limit the
> impact:

Since we're talking about a pretty specific connectivity problem (you
either see the US or you don't), could the problem be addressed by using a
sacrificial Squid box of some sort? Perhaps you could route all requests
to US sites through a parent proxy who's entire job is to proxy for
requests to US sites. Then if the US goes away, _that_ proxy gets wedged
up, but the others should be able to continue proxying non-US sites.

[If that won't work, let me know why not, because it would indicate a
misunderstanding on my part about how Squid clusters might work,]

-- 
scott hess <scott@doubleu.com> (408) 739-8858 http://www.doubleu.com/
<Favorite unused computer book title: The Compleat Demystified Idiots
  Guide to the Zen of Dummies in a Nutshell in Seven Days, Unleashed>