As it seems to be working it is like this:
Client -> squid <-> http server
Which means the Squid always connects to the http server to get the files.
My question is, I would be more scalable if the
clients would be able to connect induvidually to the http server. like this:
client <-> squid
(squid modifies the http url for flv to http server instead
of serving it self)
client <-> http server (to download content)
Actually, videocache already works like this. Squid issues a 'HTTP 303 See Other' response which directs the user's browser to connect directly to the given URL. This provides the 'client <-> http server' connection you're looking for.